CN107578368A

CN107578368A - Multi-object tracking method and panorama camera applied to panoramic video

Info

Publication number: CN107578368A
Application number: CN201710773980.2A
Authority: CN
Inventors: 黄敏; 陈卓
Original assignee: Chengdu Guan Chong Chong Yu Technology Co Ltd
Current assignee: Chengdu Guan Chong Chong Yu Technology Co Ltd
Priority date: 2017-08-31
Filing date: 2017-08-31
Publication date: 2018-01-12

Abstract

The invention discloses a kind of multi-object tracking method and panorama camera applied to panoramic video, certain the frame panoramic picture chosen in panoramic video is initial frame panoramic picture, N number of projected image is chosen in every frame panoramic picture from the initial frame panoramic picture and afterwards, M tracking target is chosen in the N number of projected image chosen from the initial frame panoramic picture；Using the initial frame panoramic picture as the 1st frame panoramic picture, the every frame panoramic picture execution following steps being directed to successively after the 1st frame panoramic picture, it is tracked using each tracking target in N number of projected image in the i-th frame panoramic picture, obtains tracking successfully at least one tracking target in N number of projected image in (i+1) frame panoramic picture.Multi-object tracking method and panorama camera provided by the invention applied to panoramic video, multiple tracking targets effectively can be accurately tracked, and the real-time for tracking multiple tracking targets can be improved.

Description

Multi-object tracking method and panorama camera applied to panoramic video

Technical field

The present invention relates to panoramic video technical field, more particularly to a kind of multi-object tracking method applied to panoramic video And panorama camera.

Background technology

Panoramic video is made up of the panoramic image sequence associated with some having times, can be obtained from video true Panorama scene, and positioning video frame can be carried out with passage time.Shoot the global fortune of the cube panoramic video of static scene Move generally based on horizontal movement, cover 360 degree of scene domains, for the panoramic picture of continuous multiple frames, target object Adjacent surface can be moved to from a face of cube panorama.Because it has dynamic background, be not suitable for the method using background subtraction Carry out the detect and track of dynamic object, it is therefore desirable to design the target following in a kind of cube panoramic video rapidly and efficiently Method.

The method that prior art tracks object in panoramic video, generally according to user viewing angle by panoramic video figure As projecting into projection view, and as the input picture of target following.The model of target object is established during tracking, and The probability for tracking target and appearing in different zones is established according to front and rear frame, the new position of target is used as using the position of greater probability； It is not in missing inspection that premise using the above method, which is, and its direction of interest is going out necessarily comprising the target for needing to track In the case of existing missing inspection, the accuracy that it is tracked is extremely low, and when needing to track multiple targets, there is significantly carrying for missing inspection in it Height, the problem of multiple target accuracys of panoramic video tracking are relatively low be present so as to cause to occur existing method.

And use the above method, it is necessary first to first obtain full-view video image and projected to obtain the throwing of preset field of view again Shadow image, then carry out the probability that front and rear frame tracking target appears in different zones, determine to track target with the size of probability New position, its calculation procedure is more and amount of calculation is larger, and when carrying out multiple target tracking, its amount of calculation can be multiplied, Consequently leads to panoramic video tracking it is less efficient the problem of.

The content of the invention

, can be effectively accurate the invention provides a kind of multi-object tracking method and panorama camera applied to panoramic video Multiple tracking targets are tracked, and the real-time for tracking multiple tracking targets can be improved.

The application first aspect provides a kind of multi-object tracking method applied to panoramic video, including：

Choose panoramic video in certain frame panoramic picture be initial frame panoramic picture, from the initial frame panoramic picture and it N number of projected image is chosen in every frame panoramic picture afterwards, wherein, each projected image is the part figure in corresponding panoramic picture The projection of picture, N are the integer not less than 2；

M tracking target is chosen in the N number of projected image chosen from the initial frame panoramic picture, wherein M is not less than 1 Integer；

It is complete for every frame after the 1st frame panoramic picture successively using the initial frame panoramic picture as the 1st frame panoramic picture Scape image performs following steps, wherein, i gets s from 1 successively, and s is the integer more than 1：

It is tracked, is obtained (i+1) using each tracking target in N number of projected image in the i-th frame panoramic picture Successfully at least one tracking target is tracked in N number of projected image in frame panoramic picture.

It is described from every in panoramic video with reference in a first aspect, in the first possible embodiment of first aspect N number of projected image is chosen in frame panoramic picture, is specifically included：

From north and south the two poles of the earth image in every frame panoramic picture remove after remaining panoramic picture in choose N number of perspective view Picture.

With reference to the possible embodiment of the first of first aspect or first aspect, second in first aspect is possible In embodiment, N number of projected image is chosen in every frame panoramic picture from panoramic video, is specifically included：

According to the corresponding relation between panoramic picture and the pixel of projected image, from every frame panorama sketch in panoramic video N number of projected image is chosen as in.

With reference in a first aspect, in the third possible embodiment of first aspect, from the initial frame panoramic picture M tracking target is chosen in the N number of projected image chosen, is specifically included：

According to preparatory condition, chosen in the N number of projected image extracted from the initial frame panoramic picture the M with Track target.

With reference to the third possible embodiment of first aspect, in the 4th kind of possible embodiment of first aspect In, according to preparatory condition, the M tracking mesh is chosen in the N number of projected image extracted from the initial frame panoramic picture Mark, is specifically included：

Chosen according to the preparatory condition, in the N number of projected image extracted from the initial frame panoramic picture all Track target；

All tracking targets are transformed into the same coordinate system, obtain has overlapping region in the same coordinate system Target is tracked to collection；

A tracking target of each tracking target centering is got rid of from all tracking targets, by remaining whole Target is tracked as described M tracking target.

With reference to the 4th kind of possible embodiment of first aspect, in the 5th kind of possible embodiment of first aspect In, a tracking target of each tracking target centering is got rid of from all tracking targets, is all tracked remaining Target specifically includes as described M tracking target:

Obtain the confidence level that each tracking target centering each tracks target；

Get rid of the small tracking target of each tracking target centering confidence level, obtain remaining all tracking targets and by its As described M tracking target.

With reference in a first aspect, in the 6th kind of possible embodiment of first aspect, from all tracking targets A tracking target of each tracking target centering is got rid of, the remaining target that all tracks is tracked into target as described M, Specifically include:

With reference in a first aspect, in the 7th kind of possible embodiment of first aspect, using in the i-th frame panoramic picture Each tracking target in N number of projected image is tracked, and is specifically included：

Create with per frame panoramic picture in N number of projected image N number of tracker correspondingly；

According to the corresponding relation of tracking target and projected image, each tracking target in the i-th frame panoramic picture is added It is tracked into N number of tracker in corresponding tracker.

With reference in a first aspect, in the 8th kind of possible embodiment of first aspect, in using the i-th frame panoramic picture N number of projected image in each tracking target when being tracked, methods described also includes：

If tracking successfully tracking target is not present in N number of projected image in (i+1) frame panoramic picture, from (i + 1) at least one tracking target is chosen in N number of projected image in frame panoramic picture.

With reference in a first aspect, in the 9th kind of possible embodiment of first aspect, from the initial frame panorama sketch After M tracking target is chosen in N number of projected image that picture is chosen, methods described is further comprising the steps of：

Using the initial frame panoramic picture as starting point, at interval of N number of corresponding to k frames panoramic picture from respective frame panoramic picture At least one tracking target is chosen in projected image, wherein, k is the integer not less than 2；

Within the cycle of each interval k frame panoramic pictures, the respective frame panoramic picture of at least one tracking target will be chosen As the 1st frame panoramic picture, following steps are performed for every frame panoramic picture after the 1st frame panoramic picture successively, wherein, i Successively k+1 is got from 1；

It is possible with reference to first aspect or the 9th kind of possible embodiment of first aspect, the tenth kind in first aspect In embodiment, successfully at least one tracking target is tracked in N number of projected image in obtaining (i+1) frame panoramic picture Afterwards, methods described also includes：

Successfully at least one tracking target will be tracked in N number of projected image in (i+1) frame panoramic picture to be shown in In (i+1) frame panoramic picture or (i+2) frame panoramic picture.

With reference to first aspect or the 9th kind of possible embodiment of first aspect, the tenth in first aspect is a kind of possible Embodiment in, in N number of projected image in obtaining (i+1) frame panoramic picture track successfully at least one tracking mesh Timestamp, methods described are further comprising the steps of：

Whether each tracking target in N number of projected image in detection (i+1) frame panoramic picture tracks successfully；

In one in often detecting (i+1) frame panoramic picture tracking target following failure, according to projected image it Between syntople, obtain with the tracking failure tracking target where projected image abut neighboring projection images；

Utilize progress target inspection in neighboring projection images of the tracking target of tracking failure in (i+2) frame panoramic picture Survey, obtain testing result corresponding to the tracking target of tracking failure；

It is successfully every that testing result sign detection is obtained in testing result corresponding to tracking target from all tracking failures Individual tracking target, at least one tracking target that tracking target obtains with tracking successful each will be detected as (i+2) frame All tracking targets in N number of projected image in panoramic picture.

With reference to a kind of the tenth possible embodiment of first aspect, in the 12nd kind of possible embodiment party of first aspect In formula, overlapping region be present in each projected image and at least one projected image, and often detect (i+1) frame panoramic picture In one tracking target following failure when, methods described is further comprising the steps of：

Whether the tracking target for detecting tracking failure is located in the overlapping region between two projected images；

If detecting, the tracking target of tracking failure is not located in the overlapping region between two projected images, abandons The tracking target of tracking failure；

If detecting the tracking target of tracking failure in the overlapping region between two projected images, obtain and Projected image where the tracking target of tracking failure has the specific projection image of overlapping region；

The tracking target to be failed using tracking carries out mesh in the specific projection image in (i+2) frame panoramic picture Mark detection, obtains testing result corresponding to the tracking target of tracking failure；

According to the second aspect of the invention, there is provided a kind of panorama camera, including readable storage medium storing program for executing, the readable storage medium Had program stored therein in matter, following steps are realized when described program is executed by processor：

With reference to second aspect, in the first possible embodiment of second aspect, described program is executed by processor When choosing the step of N number of projected image from every frame panoramic picture in panoramic video, specifically include following steps：From every frame North and south the two poles of the earth image in panoramic picture remove after remaining panoramic picture in choose N number of projected image.

With reference to the possible embodiment of the first of second aspect or second aspect, second in second aspect is possible In embodiment, N number of projected image is chosen from every frame panoramic picture in panoramic video when described program is executed by processor Step when, specifically include following steps：

With reference to second aspect, in the third possible embodiment of second aspect, described program is executed by processor When from the initial frame panoramic picture choose N number of projected image in choose M track target step when, specifically include following Step：

With reference to the third possible embodiment of second aspect, in the 4th kind of possible embodiment of second aspect In, according to preparatory condition when described program is executed by processor, the N number of perspective view extracted from the initial frame panoramic picture When the step of the M tracking target is chosen as in, following steps are specifically included：

Beneficial effects of the present invention are as follows：

Because the embodiment of the present application is entered using each tracking target in N number of projected image in the i-th frame panoramic picture Line trace, obtain tracking successfully at least one tracking target in N number of projected image in (i+1) frame panoramic picture；So that Each tracking target is carrying out target following in corresponding projected image；The embodiment of the present invention is without N number of wide angle picture is turned Change panoramic picture into；And use track algorithm amount of calculation also will be significantly lower than prior art use track algorithm determine with The amount of calculation of the new position of track target.Therefore, compared with prior art, when carrying out multiple target tracking its calculation procedure reduce and Amount of calculation reduces, thus promotes multiple target tracking efficiency in panoramic video to improve, and effectively increases multiple mesh in panoramic video Mark the real-time of tracking.

Further, the projection of the parts of images in the panoramic picture corresponding to each projected image is so that projection The distortion of image is minimum or is not present；And distort it is minimum or in the absence of the projected image of distortion in carry out target following when, its The accuracy of tracking can greatly improve, so as to effectively improve the accuracy of tracking.

Brief description of the drawings

Fig. 1 is the first flow chart for the multi-object tracking method for being applied to panoramic video in the embodiment of the present invention；

Fig. 2 is the corresponding diagram of panoramic picture and projected image in the embodiment of the present invention；

Fig. 3 is the method flow diagram for choosing M tracking target in the embodiment of the present invention from N number of projected image；

Fig. 4 is second of flow chart of the multi-object tracking method for being applied to panoramic video in the embodiment of the present invention；

Fig. 5 is the flow chart of processing method when target following failure is tracked in the embodiment of the present invention.

Embodiment

As shown in figure 1, providing a kind of multi-object tracking method applied to panoramic video in first aspect present invention, wrap Include following steps：

S101, certain the frame panoramic picture chosen in panoramic video are initial frame panoramic picture, from the initial frame panorama sketch N number of projected image is chosen in picture and every frame panoramic picture afterwards, wherein, each projected image is in corresponding panoramic picture The projection of parts of images, N are the integer not less than 2；

M tracking target is chosen in S102, the N number of projected image chosen from the initial frame panoramic picture, wherein, M is Integer not less than 1；

S103, using the initial frame panoramic picture as the 1st frame panoramic picture, successively for the 1st frame panoramic picture after Following steps are performed per frame panoramic picture, wherein, i gets s from 1 successively, and s is the integer more than 1：

S1031, it is tracked using each tracking target in N number of projected image in the i-th frame panoramic picture, obtains (i+1) successfully at least one tracking target is tracked in N number of projected image in frame panoramic picture.

Wherein, in step S101, panoramic video can be recorded by multiple wide-angle lens.Passing through multiple wide-angle lens During recording panoramic video, multiple wide angle pictures are gathered using multiple wide-angle lens each moment, then can be real-time or non- Multiple wide angle pictures that each moment is gathered in real time are spliced into a frame panoramic picture；Finally sequentially in time to being spliced into All panoramic pictures be ranked up, obtain panoramic video.

In the embodiment of the present invention, the multiple wide-angle lens is arranged in a panorama camera or the multiple Wide-angle lens connects same main frame, and the application is not specifically limited.Lower mask body is arranged on one with the multiple wide-angle lens Illustrated exemplified by panorama camera.

In the embodiment of the present invention, the multiple wide-angle lens is by common wide-angle lens, bugeye lens and fish eye lens In one or more camera lenses composition, the application is not restricted.

In the embodiment of the present invention, it can be spliced into by multiple wide angle pictures that merging algorithm for images gathers each moment One frame panoramic picture.Described image stitching algorithm for example can be related to feature based based on the related stitching algorithm in region Joining method etc..If not specifying, the panoramic picture described in the present invention is cylinder panoramic image.

In specific implementation process, the initial frame panoramic picture is chosen first, then again from the initial frame panorama sketch N number of projected image is chosen in picture and every frame panoramic picture afterwards.When choosing N number of projected image from every frame panoramic picture, N number of projection can be chosen from every frame panoramic picture according to the corresponding relation between panoramic picture and the pixel of projected image Image.

Specifically, can be according to throwing when obtaining the corresponding relation between panoramic picture and the pixel of projected image The division rule of shadow image, obtain the corresponding relation between panoramic picture and the pixel of projected image.The division rule can To be the division rule based on default longitude or default latitude or based on default longitude and latitude, drawing based on preset field of view can also be Then etc., the application is not specifically limited divider.

Specifically, when the division rule is the division rule based on preset field of view, the water of projected image can first be set Rink corner and the vertical angle of visual field are looked squarely, and calculates the subpoint of projection window；According to subpoint, horizontal field of view angle and the vertical angle of visual field Choose N number of projected image, wherein it is possible to control panoramic picture each pixel can with least one projected image Pixel is corresponding；It can also control most of pixel in panoramic picture can be with the pixel at least one projected image It is corresponding.

Specifically, panoramic picture is mapped to spherical diagram picture first, one or more throw then is chosen in spherical diagram picture Shadow point, then choose N number of projected image of projection from spherical diagram picture based on division rule again.In this way, according to each perspective view The corresponding relation of picture, spherical diagram picture and panoramic picture, can obtain each pixel and panoramic picture in each projected image In pixel corresponding relation；Then each pixel in each projected image and the pixel in panoramic picture Corresponding relation, obtain N number of projected image corresponding to every frame panoramic picture., can be with if N number of projected image is represented with ProjImg Determine that the corresponding relation between panoramic picture and the pixel of N number of projected image can use { ProjImg->PanoImg } represent.

For example, a pixels in panoramic picture are corresponding with the c pixels in b projected images, the a1 pixels in panoramic picture Point is corresponding with the c1 pixels in b1 projected images, then can determine { ProjImg->PanoImg } include corresponding relation (a-> (b, c)) and (a1->(b1, c1)).Usual such a mode, obtain in b and b1 projected images in each pixel and panoramic picture Pixel corresponding relation；According to obtained corresponding relation, the pixel value of the pixel of panoramic picture is filled in b and b1 In, and then obtain b and b1.

In actual applications, as shown in Fig. 2 panoramic picture is mapped to spherical diagram as 20, then selected in comprehensive image 20 Take subpoint 200, wherein, based on division rule from spherical diagram as 20 in choose projected image 30, projected image 30 and spherical diagram This four pixels of pixel 201, pixel 202, pixel 203 and pixel 204 are corresponding on picture 20；Based on choosing projection The same mode of image 30, remaining all projected images are chosen from panoramic picture 20.

Wherein, compared with spherical diagram spherical area as corresponding to aforementioned four pixel in 20, it distorts projected image 30 Smaller or be not present, making it, data are more accurate when projected image 30 is tracked, and promote the accuracy of target following to improve.

The present invention the first embodiment in, due to the embodiment of the present invention need carry out target following, therefore, it is necessary to Ensure that each projected image has overlapping region with least one projected image, tracking target occur from a perspective view to reduce As there is the probability of situation about disappearing when entering another projected image.If each projected image is not deposited with any one projected image In overlapping region, when tracking target and entering from a projected image in another projected image, due in the absence of overlapping region, So as to which the problem of tracking target disappears from projected image occur；And deposited in each projected image and at least one projected image In overlapping region so that there is very maximum probability to enter another projected image from a projected image by overlapping region in target. It so, it is possible effectively to reduce when target enters another projected image from a projected image and the probability of situation about disappearing occur, carry The accuracy of high target following.

Specifically, the Duplication between projected image adjacent two-by-two in N number of projected image within a preset range, example As the Duplication can be not less than the 1% of projected image size and not higher than the 30% of projected image, now, the default model Enclose for 1%-30%.Preferably, it is not less than perspective view between the Duplication between image adjacent two-by-two in N number of projected image 2% as size and 15% not higher than projected image, now, the preset range is 2%-15%.In the embodiment of the present invention, Due to having Duplication between projected image adjacent two-by-two, in this way, each projected image and other perspective views can also be recorded The syntople of picture.

For example, the N number of projected image chosen in panoramic picture is b projected images, b1 projected images, b2 projected images and b3 Projected image.If b projected images only have overlapping region with b1 projected images, the syntople of b projected images is (b-> b1)；If b1 projected images have overlapping region with b and b2 projected images, the syntople of b1 projected images is (b1->(b, b2))；If b2 projected images have overlapping region with b1 and b3 projected images, the syntople of b2 projected images is (b2-> (b1, b3))；If b3 projected images have overlapping region with b2 projected images, the syntople of b3 projected images is (b3-> b2)。

In an alternative embodiment of the invention, panorama camera uses the visual angle relatively uniform with human eye, in most cases, The probability that north and south the two poles of the earth image in panoramic picture includes tracking target is smaller, and the distortion of north and south the two poles of the earth image causes compared with conference The problem of accuracy that appearance tracks is poor.In this way, for boosting algorithm speed and the accuracy of tracking, can be from every frame panorama North and south the two poles of the earth image in image remove after remaining panoramic picture in choose N number of projected image, with improve tracking efficiency And accuracy.

More preferably, the loop graph picture of equator one that can only retain panoramic picture carries out target following, the equator side of panoramic picture To totally 360 degree, it is assumed that equator content is split into n projected image, then the longitude range that each projected image includes is 360/n+2d, d are the part that adjacent two images overlap, generally, 5≤d≤360*0.25/n, neighboring projection images Center latitude interval 360/n, the latitude span of each projected image is-e~+e；Wherein, e is latitude value, its value Scope is 35~45.

Next step S102 is performed, in this step, is extracted by step S101 in the initial frame panoramic picture N number of projected image after, according to preparatory condition, selected in the N number of projected image extracted from the initial frame panoramic picture Take the M tracking target.Wherein, M is the integer not less than 1；Preferably, M is integer not less than 2, for example, 2,3,4 Deng.

In the embodiment of the present invention, the preparatory condition can be that user voluntarily chooses or pass through target detection What method detected, it can also be what is traced into by method for tracking target, can also use in above-mentioned three kinds of methods at least What two methods obtained.The object detection method can be method for detecting human face, the depth based on Region Proposal Algorithm of target detection and biometric discrimination method etc. are practised, the application is not specifically limited.Further, the method for tracking target can With the tracking based on active profile, the tracking of feature based, the tracking based on region and based on model with Track method etc..

Specifically, when choosing the M tracking target from N number of projected image by object detection method, pass through Object detection method detects to each projected image in N number of projected image, will be directed to each projected image and detect All testing results arrived are as described M tracking target；And the M tracking target and N number of projected image will be recorded Corresponding relation.

Specifically, can be by building when recording the corresponding relation of the M tracking target and N number of projected image Vertical index marker marks each projected image, and the corresponding relation that then can record each tracking target and index marker is remembered The corresponding relation of each tracking target of record and N number of projected image.

In the embodiment of the present invention, the index marker of each projected image is unique, is prevented because index marker is not unique Cause the incorrect problem of corresponding relation for tracking target and projected image occur.Further, from the initial frame panorama When the M tracking target is chosen in the N number of projected image extracted in image, it can parallel perform, can also perform successively, The application is not specifically limited.

If for example, N=4, the first projected image, the second projected image, the 3rd projected image in 4 projected images and The index marker that 4th projected image is corresponding in turn to is T1, T2, T3 and T4；And by object detection method to 4 perspective views As carrying out target detection, the F1 targets for obtaining detecting from first projected image, examined from second projected image The F2 targets and F3 targets that measure, the F4 targets detected from the 4th projected image.It follows that 4 projected images Corresponding relation with tracking target includes { (F1->T1), ((F2, F3)->) and (F4- T2>T4)}.Lower mask body is with T1, T2, T3 Represent that the first projected image, the second projected image, the 3rd projected image and the 4th projected image are illustrated respectively with T4.

In another embodiment of the invention, it is described according to preparatory condition, extracted from the initial frame panoramic picture N number of projected image in choose it is described M tracking target, can also comprise the following steps, as shown in Figure 3：

S301, according to the preparatory condition, chosen in the N number of projected image extracted from the initial frame panoramic picture All tracking targets；

S302, by it is described it is all tracking targets be transformed into the same coordinate system, obtain in the same coordinate system have it is overlapping The tracking target in region is to collection；

S303, each tracking target for tracking target centering is got rid of from all tracking targets, by remainder Whole tracking targets track target as described M.

Wherein, in step S301, according to the preparatory condition, extracted from the initial frame panoramic picture each Tracking target is chosen in projected image, the tracking target chosen from each projected image is obtained, that is, obtains all tracking Target.

If for example, N number of projected image in the initial frame panoramic picture is T1, T2, T3 and T4, according to Face datection Algorithm detects to T1, T2, T3 and T4 respectively；Detection obtains the human face target R3 in human face target R1 and R2, T2 in T1 Human face target is not detected by with the human face target R5 and T4 in R4, T3, that is, is obtained all in the initial frame panoramic picture Tracking target be R1, R2, R3, R4 and R5, wherein the corresponding relation of 4 projected images and tracking target including ((R1, R2)-> T1), ((R3, R4)->) and (R5- T2>T3)}.

Next step S302 is performed, after all tracking targets are transformed into the same coordinate system, can incited somebody to action All tracking targets are transformed into after the same coordinate system, will transition to every two in all tracking targets in the same coordinate system Individual tracking target is contrasted；All tracking targets that overlapping region be present are contrasted to forming the tracking target to collection.

It is of course also possible to according to the syntople and track target of each projected image of record and other projected images with throwing The corresponding relation of shadow image, the coordinate of the tracking target in the projected image adjacent two-by-two after conversion is compared；Contrast Go out in all adjacent projected images two-by-two to exist the tracking target of overlapping region to forming the tracking target to collection.In this way, Compared with it will transition in all tracking targets in the same coordinate system each two tracking target and carry out contrast, what it was contrasted The quantity of each two tracking target tails off, and can reduce amount of calculation, improve computational efficiency.

For example, if T1, T2, T3 and T4 syntople are { (T1->T2), (T2->(T1, T3)), (T3->(T2, T4)), (T4->T3) }, after all tracking target R1, R2, R3, R4 and R5 are obtained；According to the syntople, by R3 and R4 successively It is compared with R1, R2 and R5；If R3 and R2 have an overlapping region, and R4 and R5 has overlapping region, it is determined that it is described with Track target is { (R2, R3), (R4, R5) } to collection.

Next step S303 is performed, can randomly choose and get rid of each tracking target from all tracking targets One tracking target of centering, obtain remaining all tracking targets and as described M tracking target.

In an alternative embodiment of the invention, in order to improve the degree of accuracy of the tracking target in the M tracking target, also may be used To obtain the confidence level that each tracking target centering each tracks target；Get rid of it is each tracking target centering confidence level it is small with Track target, obtain remaining all tracking targets and as described M tracking target.So so that the M tracking mesh The big tracking target of each tracking target centering confidence level is remained in mark, and then can be improved in the M tracking target Track the degree of accuracy of target.

Specifically, if obtain track target when obtained it is each tracking target confidence level, get rid of it is each with During the small tracking target of track target centering confidence level, it can be carried out according to two confidence levels of each tracking target pair obtained Comparison.For example, when obtaining each tracking target by object detection method and/or method for tracking target, target inspection Survey method and/or method for tracking target can calculate the confidence level of each tracking target automatically.In this way, getting rid of each tracking During the small tracking target of target centering confidence level, it can be compared according to two confidence levels of each tracking target pair obtained To.

Further, it is each getting rid of if not obtaining the confidence level of each tracking target when obtaining and tracking target Before tracking the small tracking target of target centering confidence level, the confidence of each tracking target can be obtained according to certainty factor algebra Degree, is then compared according to two confidence levels of each tracking target pair of acquisition.Wherein, the certainty factor algebra for example may be used To be confidential interval computational methods, normal distribution method etc..

Further, during according to certainty factor algebra obtaining the confidence level of tracking target, can also be calculated according only to confidence level Method obtains the confidence level that all trace point centerings each track target, without obtaining the confidence level of each tracking target, with Amount of calculation is reduced, and then improves and calculates effect.

For example, it is R1, R2, R3, R4 and R5 that if all tracking targets corresponding to T1, T2, T3 and T4, which are all tracking targets, And corresponding tracking target is { (R2, R3), (R4, R5) } to collection, if confidence level corresponding to R2 is confidence level corresponding to 0.72, R3 It is that confidence level corresponding to 0.78, R5 is 0.75 for confidence level corresponding to 0.85, R4, due to 0.72<0.85 and 0.78>0.75, then The small R2 of confidence level is got rid of from R2 and R3 and the small R5 of confidence level is got rid of from R4 and R5, obtains remaining all tracking Target is R1, R3 and R4, i.e. R1, R3 and R4 is described M tracking target, now M=3.

Next step S103 is performed, performs step for every frame panoramic picture after the 1st frame panoramic picture successively S1031。

Wherein, in step S1031, entered using each tracking target in N number of projected image in the i-th frame panoramic picture During line trace, corresponding projected image of each tracking target in (i+1) frame panoramic picture in the i-th frame panoramic picture is entered Line trace, so as to obtain that successfully at least one tracking target is tracked in N number of projected image in (i+1) frame panoramic picture.

During the present invention is implemented, because N number of projected image in every frame panoramic picture is chosen according to the preparatory condition , and correspond to the preparatory condition all same per frame panoramic picture；It follows that per N number of projected image in frame panoramic picture It is identical with the visual field corresponding to N number of projected image in other any one frame panoramic pictures；Therefore, former frame can be caused It is tracked in corresponding projected image of each tracking target in next frame panoramic picture in panoramic picture.

If for example, the projected image in the initial frame panoramic picture is T1, T2, T3 and T4,4 projected images are with tracking The corresponding relation of target for ((R1, R2)->T1), ((R3, R4)->) and (R5- T2>T3)}；And corresponding M tracking target For R1, R3 and R4；If the projected image in next frame panoramic picture is T11, T21, T31 and T41, wherein, T1 is corresponding with T11, T21 is corresponding with T2, and T31 is corresponding with T3, and T41 is corresponding with T4；Further according to the corresponding relation of projected image and tracking target, mesh is utilized R1 is tracked by mark track algorithm in T11, is tracked R3 and R4 in T21 using target tracking algorism.

Specifically, because i gets s from 1 successively, in i=1, then tracked using M in the 1st frame panoramic picture in target Each tracking target be tracked；And due to step S1031 of every execution, (i+1) frame panoramic picture can be obtained In N number of projected image in track successfully at least one tracking target so that when i gets s from 1 successively, can utilize obtain (i+1) frame panoramic picture in track successfully each tracking target and perform step S1031.For example, in i=1, then utilize Each tracking target in i-th frame panoramic picture in M tracking target is tracked, and obtains N number of throwing in the 2nd frame panoramic picture Successfully at least one tracking target is tracked in shadow image；In i=2, then using the tracking in the 2nd obtained frame panoramic picture into Each tracking target of work(is tracked, and then i 3 gets s always successively.

In another embodiment of the present invention, it can also create and the one-to-one N of N number of projected image in every frame panoramic picture Individual tracker；Further according to the corresponding relation of tracking target and projected image, each tracking target in the i-th frame panoramic picture is added It is added in N number of tracker and is tracked in corresponding tracker.

During the present invention is implemented, the tracker is the trace model calculated using target tracking algorism, such as basis The parameter read-in trace model of a tracking clarification of objective point of some projected image, then described in former frame panoramic picture Tracker carries out computing to be tracked to the parameter of write-in automatically.Wherein, the target tracking algorism includes being based on CamShift target tracking algorism, the target tracking algorism based on average drifting and Kalman filtering, based on SIFT feature mesh Mark track algorithm research and Dynamic Object Tracing Algorithm based on Kalman Prediction etc..

For example, it is T1, T2, T3 and T4 that if initial frame panoramic picture, which is N number of projected image in the 1st frame panoramic picture, G1 trackers, G2 trackers, G3 trackers and G4 trackers this four trackers are created, wherein, G1 is corresponding with T1, G2 and T2 pairs Should, G3 is corresponding with T3 and G4 and T4 is corresponding.And M tracking target be R1, R3 and R4, because R1 is corresponding with T1, R3 and T2 pairs Should be corresponding with T3 with R4, then according to tracker and the corresponding relation of projected image, R1 is added in G1, R3 is added to G2 In, and R4 is added in G3；If the projected image in the 2nd frame panoramic picture is T11, T21, T31 and T41, wherein, T1 with T11 is corresponding, T21 is corresponding with T2, and T31 is corresponding with T3, and T41 is corresponding with T4；R1 is then tracked in T11 by G1, by G2 in T21 Middle tracking R3 and R4；Obtain tracking successful R1 from the projected image in the 2nd frame panoramic picture and represented and tracked successfully with R11 R3 represented with R31；If the projected image in the 3rd frame panoramic picture is T12, T22, T32 and T42, existed using R11 by G1 It is tracked in T12 and is tracked using R31 by G2 in T22, is obtained from the projected image in the 3rd frame panoramic picture Track successful R11 and represent and track successful R31 with R12 and represented with R32.

, then can be multiple with Parallel Tracking when the tracking target in the i-th frame panoramic picture is multiple in the embodiment of the present invention Target is tracked, to shorten the tracking spent time；It is of course also possible to serially track multiple tracking targets, the application does not make to have Body limits.

In another embodiment of the present invention, each tracking target in N number of projected image in using the i-th frame panoramic picture When being tracked, if tracking successfully tracking target is not present in N number of projected image in (i+1) frame panoramic picture, from At least one tracking target is chosen in N number of projected image in (i+1) frame panoramic picture；It is of course also possible to from (i+2) frame At least one tracking target is chosen in N number of projected image in panoramic picture.Wherein, from (i+1) or (i+2) frame panorama sketch , can be according to the preparatory condition, from (i+1) or (i+ when at least one tracking target is chosen in N number of projected image as in 2) at least one tracking target is chosen in N number of projected image in frame panoramic picture.It chooses the tool of at least one tracking target The narration for step S101 is may be referred in body implementation process, it is succinct for specification, just repeat no more herein.

If for example, the projected image in the 3rd frame panoramic picture is T12, T22, T32 and T42, existed using R11 by G1 It is tracked in T12 and is tracked using R31 by G2 in T22；If R11 and R31 do not track success, according to Preparatory condition, at least one tracking target is chosen from T12, T22, T32 and T42.It is of course also possible to according to the default bar Part, it is to choose at least one tracking target in T13, T23, T33 and T43 from the projected image in the 4th frame panoramic picture.

In another embodiment of the present invention, chosen in the N number of projected image chosen from the initial frame panoramic picture M with After track target, methods described is further comprising the steps of, referring to Fig. 4：

S401, using the initial frame panoramic picture as starting point, it is corresponding from respective frame panoramic picture at interval of k frames panoramic picture N number of projected image in choose at least one tracking target, wherein, k is integer not less than 2；

S402, within the cycle of each interval k frame panoramic pictures, will choose it is at least one tracking target respective frame panorama Image performs following steps as the 1st frame panoramic picture for every frame panoramic picture after the 1st frame panoramic picture successively, its In, i gets k+1 from 1 successively：

S4021, it is tracked using each tracking target in N number of projected image in the i-th frame panoramic picture, obtains (i+1) successfully at least one tracking target is tracked in N number of projected image in frame panoramic picture.

Wherein, in step S401, it is determined that during respective frame panoramic picture, successively to each interval k frame panoramic pictures Multiframe panoramic picture in cycle carries out target detection, and certain the frame panoramic picture for detecting at least one tracking target first is made For respective frame panoramic picture.So so that respective frame panoramic picture can be opening in the cycle of some interval k frame panoramic pictures Beginning frame panoramic picture or panoramic picture afterwards.

If k=4 so that the panoramic picture that the cycle of first 4 frame panoramic picture of interval includes is the 1st, 2,3,4 and 5 frames Panoramic picture, wherein the start frame panoramic picture in the cycle of first 4 frame panoramic picture of interval is the 1st frame panoramic picture, then Successively target detection is carried out for the 1st, 2,3,4 and 5 frame panoramic pictures；If from N number of projected image corresponding to the 1st frame panoramic picture In detect at least one tracking target, it is determined that the 1st frame panoramic picture is respective frame panoramic picture, now, in interval k Respective frame panoramic picture in the cycle of frame panoramic picture is initial frame panoramic picture；Accordingly, second 4 frame panorama sketch of interval The panoramic picture that the cycle of picture includes is the 6th, 7,8,9 and 10 frame panoramic pictures, wherein the week of second 4 frame panoramic picture of interval Start frame panoramic picture in phase is the 6th frame panoramic picture, then carries out target for the 6th, 7,8,9 and 10 frame panoramic pictures successively Detection；If not detecting tracking target from the 6th frame panoramic picture, target is carried out to the 7th frame panoramic picture；If from the 7th frame panorama Image has detected at least one tracking target, it is determined that the 7th frame panoramic picture is respective frame panoramic picture, now, in interval k Respective frame panoramic picture in the cycle of frame panoramic picture is 1 frame panoramic picture after initial frame panoramic picture.

In the embodiment of the present invention, step S401 embodiment can be step S102 any one embodiment party Formula, it is succinct for specification, just repeat no more herein.

After step S401 is performed, within the cycle of each interval k frame panoramic pictures, to choose at least one tracking mesh Target respective frame panoramic picture is the 1st frame panoramic picture, then successively for every after the 1st frame panoramic picture in the cycle Frame panoramic picture performs step S4021, and wherein step S403 specific implementation process may be referred to step S1032 implementation Journey.

For example, if the panoramic picture that the cycle of second 4 frame panoramic picture of interval includes is the 6th, 7,8,9 and 10 frame panoramas Image, first the start frame panoramic picture out of second 4 frame panoramic picture of interval cycle are the throwing in the 6th frame panoramic picture Shadow image is T15, T25, T35 and T45；At this point it is possible to chosen according to method for detecting human face from T15, T25, T35 and T45 to Few tracking target；Then at least one tracking target chosen corresponding projected image in the 7th frame panoramic picture is utilized In be tracked.

Due in the present embodiment at interval of k frames panoramic picture just from N number of projected image corresponding to respective frame panoramic picture Choose at least one tracking target so that there is new tracking target to enter in scene corresponding to panoramic picture and fashionable can detect immediately To and be tracked.So, it is possible improve detect tracking target promptness and accuracy so that the accuracy of tracking and and When property can also improve therewith.

In another embodiment of the present invention, tracked in N number of projected image in obtaining (i+1) frame panoramic picture successful After at least one tracking target, successful at least one can will be tracked in N number of projected image in (i+1) frame panoramic picture Individual tracking target is shown in (i+1) frame panoramic picture or (i+2) frame panoramic picture.

Specifically, successfully at least one tracking will be being tracked in N number of projected image in (i+1) frame panoramic picture When target is shown in (i+1) frame panoramic picture or (i+2) frame panoramic picture, it is necessary first to by (i+1) frame panoramic picture Successfully at least one each tracking target tracked in target carries out Coordinate Conversion for middle tracking, then will be every after Coordinate Conversion Individual target following is shown in (i+1) frame panoramic picture or (i+2) frame panoramic picture.

Specifically, when carrying out Coordinate Conversion, can according to the pixel point correspondence in projected image and panoramic picture, Tracking is successfully each tracked into target and carries out Coordinate Conversion, so as to the Coordinate Conversion by each tracking target in projected image For the coordinate in panoramic picture；In this way, each target following after Coordinate Conversion is included in (i+1) frame panoramic picture or the (i+2) when in frame panoramic picture, its accuracy shown improves.

If for example, the projected image in the 2nd frame panoramic picture is T11, T21, T31 and T41, by G1 in T11 with Track R1, R3 and R4 is tracked in T21 by G2；Obtain tracking successful R1 use from the projected image in the 2nd frame panoramic picture R11 represents and tracks successful R3 and represented with R31, then will R11 and R31 carry out Coordinate Conversion after include in the 2nd frame panorama sketch In picture or the 3rd frame panoramic picture.

In the embodiment of the present invention, successful at least one will be being tracked in N number of projected image in (i+1) frame panoramic picture Individual tracking target is when being shown in (i+1) frame panoramic picture, by N number of projected image in the 1st frame panoramic picture at least One tracking target is shown in the 1st frame panoramic picture；It will tracked in N number of projected image in (i+1) frame panoramic picture When successfully at least one tracking target is shown in (i+2) frame panoramic picture, by N number of projection in the 1st frame panoramic picture At least one tracking target in image is shown in the 2nd frame panoramic picture；By the above-mentioned means, to prevent the 1st and 2 frames The problem of at least one tracking target in N number of projected image in panoramic picture is simultaneously displayed in the 2nd frame panoramic picture, drop It is low the problem of tracking target overlaps display occur.

In another embodiment of the present invention, each tracking target in N number of projected image in using the i-th frame panoramic picture It is tracked, when obtaining tracking successfully at least one tracking target in N number of projected image in (i+1) frame panoramic picture, Methods described is further comprising the steps of, as shown in Figure 5：

Whether each tracking target in N number of projected image in S501, detection (i+1) frame panoramic picture tracks into Work(；

S502, in one in often detecting (i+1) frame panoramic picture tracking target following failure, according to perspective view Syntople as between, obtain the neighboring projection images with the projected image adjoining where the tracking target of tracking failure；

S503, the tracking target to be failed using tracking are carried out in the neighboring projection images in (i+2) frame panoramic picture Target detection, obtain testing result corresponding to the tracking target of tracking failure；

Testing result sign is obtained in testing result corresponding to S504, the tracking target from all tracking failures to detect successfully Each tracking target, at least one tracking target that successfully each tracking target and tracking obtain will be detected as (i+ 2) all tracking targets in N number of projected image in frame panoramic picture.

Wherein, in step S501, each tracking target in N number of projected image in using the i-th frame panoramic picture When being tracked, whether each tracking target detected in the i-th frame panoramic picture tracks success；If each tracking target tracks Success, then can determine that all tracking targets in N number of projected image in (i+1) frame panoramic picture track success；With And when often detecting a tracking target following failure, perform step S502.

If for example, the projected image in the 2nd frame panoramic picture is T11, T21, T31 and T41, by G1 in T11 with Track R1, R3 and R4 is tracked in T21 by G2；Detect that R1 is tracked successfully to obtain R11 and R3 and track successfully to obtain R31, and R4 tracking failures are detected, now, perform step S502.

When often detecting a tracking target following failure, step S502 is performed, due in tracking target following failure When, tracking the tracking target of failure has projected image of the larger probability where it to be moved in the neighboring projection images； Therefore, the neighboring projection images are got by step S502, then by step S503 in (i+2) frame panoramic picture The tracking target of tracking failure is detected in neighboring projection images, obtains testing result；In this way, part tracking can be lost The tracking target lost detects again to be tracked again, and then can realize and tracking target is more accurately tracked, and is dropped The probability that low tracking target is lost.

If for example, the projected image in the 2nd frame panoramic picture is T11, T21, T31 and T41, by G1 in T11 with Track R1, R3 and R4 is tracked in T21 by G2；Detect that R1 is tracked successfully to obtain R11 and R3 and track successfully to obtain R31, and R4 tracking failures are detected, then according to syntople { (T1->T2), (T2->(T1, T3)), (T3->(T2, T4)), (T4-> T3) }, it is determined that being T1 and T3 with the neighboring projection images abutted of the T2 where R4；Then the projection in the 3rd frame panoramic picture is obtained Image is T12, T22, T32 and T42, and projected image corresponding with T1 and T3 is T12 and T32 in the 3rd frame panoramic picture, then utilizes R4 carries out target detection in T12 and T32, obtains testing result.

Wherein, if the testing result, which is characterized in T32, detects that R4 is represented with R42, and utilize R11 by G1 in T12 In be tracked and be tracked using R31 by G2 in T22, obtain from the projected image in the 3rd frame panoramic picture with The successful R11 of track represents and tracked successful R31 with R12 and represented with R32, it is determined that 4 perspective views in the 3rd frame panoramic picture All tracking targets as in are R12, R32 and R42.

Further, if the projected image in the 4th frame panoramic picture is T13, T23, T33 and T43, passed through using R12 G1 is tracked in T13, is tracked in T23 by G2 using R32 and is tracked using R42 by G3 in T33, Obtain tracking successful R12 from the projected image in the 3rd frame panoramic picture and represented with R13, track successful R32 R33 tables Show and track successfully R42 and represented with R43.In this way, the tracking target that part can be tracked to failure is detected and carried out again again Tracking, and then can realize and tracking target is more accurately tracked, reduce the probability that tracking target is lost.

In another embodiment of the invention, overlapping region be present in each projected image and at least one projected image, And when often detecting one in (i+1) frame panoramic picture tracking target following failure, methods described is further comprising the steps of：

S601, detect whether the tracking target that the tracking fails is located in the overlapping region between two projected images；

If S602, detecting that the tracking target of tracking failure is not located in the overlapping region between two projected images, Then abandon the tracking target of tracking failure；

In the embodiment of the present invention, if detecting, the tracking target of tracking failure is not located at the weight between two projected images In folded region, step S502~S504 can also be performed successively.

If S603, detecting the tracking target of tracking failure in the overlapping region between two projected images, Obtain the specific projection image that there is overlapping region with the projected image where the tracking target of tracking failure；

In the specific projection image of S604, the tracking target to be failed using tracking in (i+2) frame panoramic picture Target detection is carried out, obtains testing result corresponding to the tracking target of tracking failure；

Testing result sign is obtained in testing result corresponding to S605, the tracking target from all tracking failures to detect successfully Each tracking target, at least one tracking target that successfully each tracking target and tracking obtain will be detected as (i+ 2) all tracking targets in N number of projected image in frame panoramic picture.

Wherein, in step s 601, tracking failure can be detected according to the coordinate of the tracking target of tracking failure Track whether target is located in the overlapping region between two projected images.Certainly, in order to quickly detect the tracking fail Tracking target whether in the overlapping region between two projected images, can also be according to the tracking target of tracking failure Coordinate and projected image between syntople, detect tracking failure tracking target whether positioned at two projected images it Between overlapping region in.

Specifically, first by the tracking failure tracking target coordinate carry out Coordinate Conversion, by the tracking unsuccessfully with The Coordinate Conversion of track target is into the latitude and longitude coordinates in panoramic picture；Then the latitude and longitude coordinates that will convert into again and each projection Latitude and longitude coordinates in panoramic picture corresponding to image are compared；If the latitude and longitude coordinates being converted into and two projected images pair There is same section in the latitude and longitude coordinates answered, then can be determined that the tracking target of tracking failure between two projected images Overlapping region in, then perform step S603~S605；If the latitude and longitude coordinates being converted into are only corresponding with a projected image There is same section in latitude and longitude coordinates, then can be determined that the tracking target step of tracking failure between two projected images In overlapping region, then step S602 is performed.

Specifically, after the coordinate of tracking target of tracking failure is obtained, according to the adjacent pass between projected image System, it is determined that the projected image adjacent with projected image where the tracking target of tracking failure；Then by the tracking failure with The Coordinate Conversion of track target is into the latitude and longitude coordinates in panoramic picture, and the latitude and longitude coordinates and neighboring projection images that will convert into Latitude and longitude coordinates in corresponding panoramic picture are compared；If the latitude and longitude coordinates being converted into are corresponding with neighboring projection images There is same section in latitude and longitude coordinates, then can be determined that weight of the tracking target of tracking failure between two projected images In folded region, then step S603~S605 is performed；If the latitude and longitude coordinates being converted into longitude and latitude corresponding with neighboring projection images Same section is not present in coordinate, then can be determined that the tracking target step of tracking failure is overlapping between two projected images In region, then step S602 is performed.

Now, the latitude and longitude coordinates being converted into only need the latitude and longitude coordinates in panoramic picture corresponding with neighboring projection images It is compared, without being compared with the latitude and longitude coordinates in the panoramic picture of each projected image, its quantity contrasted subtracts It is few, promote its amount of calculation also to reduce therewith, detection efficiency is improved so as to shorten detection time.

If detecting, the tracking target of tracking failure in the overlapping region between two projected images, performs step S603, at this point it is possible to two projected images where obtaining the tracking target of tracking failure corresponding to overlapping region, place weight Two projected images one corresponding to folded region are the projected image where the tracking target of tracking failure, and another is institute Specific projection image is stated, the specific projection image can be got by such a mode.

In the embodiment of the present invention, step S604~S605 specific implementation process may be referred to step S503~S504 reality Process is applied, it is succinct for specification, just repeat no more herein.

If for example, the projected image in the 2nd frame panoramic picture is T11, T21, T31 and T41, by G1 in T11 with Track R1, R3 and R4 is tracked in T21 by G2；Detect that R1 is tracked successfully to obtain R11 and R3 and track successfully to obtain R31, and R4 tracking failures are detected, detection R4 is whether in the no overlapping region being located between two projected images.

Because R4 is located in projected image T2, and according to syntople { (T1->T2), (T2->(T1, T3)), (T3-> (T2, T4)), (T4->T3) }, it is determined that being T1 and T3 with the neighboring projection images abutted of the T2 where R4；Then by R4 coordinate Be converted into latitude and longitude coordinates, then the latitude and longitude coordinates that will convert into respectively with T1 and T3 corresponding to latitude and longitude coordinates be compared. If the latitude and longitude coordinates that R4 is converted into latitude and longitude coordinates corresponding with T3 have same section, it can determine that R4 is located at T2 and T3 Between overlapping region, and because the projected image where R4 is T2, it follows that specific projection image corresponding to R4 is T3.

Wherein, it is determined that specific projection image corresponding to R4 be T3 after, if the projected image in the 3rd frame panoramic picture is T12, T22, T32 and T42, projected image corresponding with T3 is T32 in the 3rd frame panoramic picture, then mesh is carried out in T32 using R4 Mark detection, obtains testing result.

Secondly, if the testing result, which is characterized in T32, detects that R4 is represented with R42, and utilize R11 by G1 in T12 In be tracked and be tracked using R31 by G2 in T22, obtain from the projected image in the 3rd frame panoramic picture with The successful R11 of track represents and tracked successful R31 with R12 and represented with R32, it is determined that 4 perspective views in the 3rd frame panoramic picture All tracking targets as in are R12, R32 and R42.

During the present invention is implemented, whether the tracking target by detecting tracking failure is located at the weight between two projected images Testing result in folded region alternatively condition, the tracking target of tracking failure of target detection need not be carried out by removing part, Tracking target without all tracking failures will carry out target detection, and tracking is improved so as to effectively reduce amount of calculation Efficiency.Moreover, after the less probability of tracking target that the tracking that the part removed need not carry out target detection fails appears in In continuous frame panoramic picture, the accuracy for the tracking target for needing the tracking failure for carrying out target detection is improved.

Based on the method identical technical concept provided with first aspect present invention, second aspect of the present invention additionally provides one Panorama camera, including readable storage medium storing program for executing are planted, is had program stored therein on the readable storage medium storing program for executing, described program is executed by processor Shi Shixian following steps：

Choose panoramic video in certain frame panoramic picture be initial frame panoramic picture, from the initial frame panoramic picture and it N number of projected image is chosen in every frame panoramic picture afterwards,；

Preferably, N number of projection is chosen when described program is executed by processor from every frame panoramic picture in panoramic video During the step of image, following steps are specifically included：From north and south the two poles of the earth image in every frame panoramic picture remove after residue it is complete N number of projected image is chosen in scape image.

Preferably, N number of projection is chosen when described program is executed by processor from every frame panoramic picture in panoramic video During the step of image, following steps are specifically included：

Preferably, in the N number of projected image chosen when described program is executed by processor from the initial frame panoramic picture When choosing the M step for tracking target, following steps are specifically included：

Preferably, according to preparatory condition when described program is executed by processor, extracted from the initial frame panoramic picture When the step of the M tracking target is chosen in the N number of projected image gone out, following steps are specifically included：

Preferably, each tracking target pair is got rid of when described program is executed by processor from all tracking targets In one tracking target, using it is remaining all track targets as described M track target step when, specifically include following Step：

Preferably, a tracking target of each tracking target centering is got rid of from all tracking targets, by more than Under whole tracking targets track target as described M, specifically include:

Preferably, using each in N number of projected image in the i-th frame panoramic picture when described program is executed by processor During the step that tracking target is tracked, following steps are specifically included：

Preferably, it is every in N number of projected image in using the i-th frame panoramic picture when described program is executed by processor It is further comprising the steps of when individual tracking target is tracked：

Preferably, after M tracking target is chosen in the N number of projected image chosen from the initial frame panoramic picture, Also realize following steps：

Preferably, when described program is executed by processor in N number of projected image in obtaining (i+1) frame panoramic picture Tracking successfully after at least one tracking target, also realizes following steps：

Preferably, when described program is executed by processor in N number of projected image in obtaining (i+1) frame panoramic picture When tracking successfully at least one tracking target, following steps are also realized：

Preferably, exist when described program is executed by processor with least one projected image in each projected image overlapping Region, and often detect one in (i+1) frame panoramic picture tracking target following failure when, also realize following steps：

Beneficial effects of the present invention are as follows：

All features disclosed in this specification, or disclosed all methods or during the step of, except mutually exclusive Feature and/or step beyond, can combine in any way.

Any feature disclosed in this specification (including any accessory claim, summary and accompanying drawing), except non-specifically chatting State, can alternative features equivalent by other or with similar purpose replaced.I.e., unless specifically stated otherwise, each feature It is an example in a series of equivalent or similar characteristics.

The invention is not limited in foregoing embodiment.The present invention, which expands to, any in this manual to be disclosed New feature or any new combination, and disclose any new method or process the step of or any new combination.

Claims

A kind of 1. multi-object tracking method applied to panoramic video, it is characterised in that including：

Certain the frame panoramic picture chosen in panoramic video is initial frame panoramic picture, from the initial frame panoramic picture and afterwards N number of projected image is chosen in per frame panoramic picture, wherein, each projected image is the parts of images in corresponding panoramic picture Projection, N are the integer not less than 2；

M tracking target is chosen in the N number of projected image chosen from the initial frame panoramic picture, wherein M is whole not less than 1 Number；

Using the initial frame panoramic picture as the 1st frame panoramic picture, successively for every frame panorama sketch after the 1st frame panoramic picture As performing following steps, wherein, i gets s from 1 successively, and s is the integer more than 1：

It is tracked using each tracking target in N number of projected image in the i-th frame panoramic picture, it is complete obtains (i+1) frame Successfully at least one tracking target is tracked in N number of projected image in scape image.
2. the method as described in claim 1, it is characterised in that choose N in every frame panoramic picture from panoramic video Individual projected image, is specifically included：

From north and south the two poles of the earth image in every frame panoramic picture remove after remaining panoramic picture in choose N number of projected image.
3. method as claimed in claim 1 or 2, it is characterised in that selected in every frame panoramic picture from panoramic video N number of projected image is taken, is specifically included：

According to the corresponding relation between panoramic picture and the pixel of projected image, from every frame panoramic picture in panoramic video Choose N number of projected image.
4. the method as described in claim 1, it is characterised in that the N number of projection chosen from the initial frame panoramic picture M tracking target is chosen in image, is specifically included：

According to preparatory condition, the M tracking mesh is chosen in the N number of projected image extracted from the initial frame panoramic picture Mark.
5. method as claimed in claim 4, it is characterised in that it is described according to preparatory condition, from the initial frame panoramic picture In choose the M tracking target in N number of projected image for extracting, specifically include：

According to the preparatory condition, all tracking are chosen in the N number of projected image extracted from the initial frame panoramic picture Target；

All tracking targets are transformed into the same coordinate system, obtain the tracking in the same coordinate system with overlapping region Target is to collection；

A tracking target of each tracking target centering is got rid of from all tracking targets, is all tracked remaining Target is as described M tracking target.
6. method as claimed in claim 5, it is characterised in that described to get rid of each tracking from all tracking targets One tracking target of target centering, using the remaining target that all tracks as described M tracking target, specifically include:

Obtain the confidence level that each tracking target centering each tracks target；

Get rid of the small tracking target of each tracking target centering confidence level, obtain remaining all tracking targets and as The M tracking target.
7. the method as described in claim 1, it is characterised in that in N number of projected image using in the i-th frame panoramic picture Each tracking target be tracked, specifically include：

Create with per frame panoramic picture in N number of projected image N number of tracker correspondingly；

According to the corresponding relation of tracking target and projected image, each tracking target in the i-th frame panoramic picture is added to institute State in N number of tracker and be tracked in corresponding tracker.
8. the method as described in claim 1, it is characterised in that in N number of projected image in using the i-th frame panoramic picture When each tracking target is tracked, methods described also includes：

If tracking successfully tracking target is not present in N number of projected image in (i+1) frame panoramic picture, from (i+1) At least one tracking target is chosen in N number of projected image in frame panoramic picture.
9. the method as described in claim 1, it is characterised in that in the N number of perspective view chosen from the initial frame panoramic picture After M tracking target is chosen as in, methods described is further comprising the steps of：

Using the initial frame panoramic picture as starting point, at interval of N number of projection corresponding to k frames panoramic picture from respective frame panoramic picture At least one tracking target is chosen in image, wherein, k is the integer not less than 2；

Within the cycle of each interval k frame panoramic pictures, will choose it is at least one tracking target respective frame panoramic picture as 1st frame panoramic picture, following steps are performed for every frame panoramic picture after the 1st frame panoramic picture successively, wherein, i is successively K+1 is got from 1；

It is tracked using each tracking target in N number of projected image in the i-th frame panoramic picture, it is complete obtains (i+1) frame Successfully at least one tracking target is tracked in N number of projected image in scape image.
10. the method as described in claim 1 or 9, it is characterised in that N number of projection in (i+1) frame panoramic picture is obtained After successfully at least one tracking target is tracked in image, methods described also includes：

Successfully at least one tracking target will be tracked in N number of projected image in (i+1) frame panoramic picture and is shown in (i+ 1) in frame panoramic picture or (i+2) frame panoramic picture.
11. the method as described in claim 1 or 9, it is characterised in that N number of projection in (i+1) frame panoramic picture is obtained When successfully at least one tracking target is tracked in image, methods described is further comprising the steps of：

Whether each tracking target in N number of projected image in detection (i+1) frame panoramic picture tracks successfully；

In a tracking target following failure in often detecting (i+1) frame panoramic picture, according between projected image Syntople, obtain the neighboring projection images with the projected image adjoining where the tracking target of tracking failure；

The tracking target to be failed using tracking carries out target detection in the neighboring projection images in (i+2) frame panoramic picture, Obtain testing result corresponding to the tracking target of tracking failure；

Obtained corresponding to tracking targets from all tracking failures in testing result testing result characterize detection successfully each with Track target, at least one tracking target that tracking target obtains with tracking successful each will be detected as (i+2) frame panorama All tracking targets in N number of projected image in image.
12. method as claimed in claim 11, it is characterised in that exist in each projected image and at least one projected image Overlapping region, and often detect one in (i+1) frame panoramic picture tracking target following failure when, methods described also includes Following steps：

Whether the tracking target for detecting tracking failure is located in the overlapping region between two projected images；

If detecting the tracking target of tracking failure in the overlapping region between two projected images, abandon this with The tracking target of track failure；

If detecting the tracking target of tracking failure in the overlapping region between two projected images, obtain be somebody's turn to do with Projected image where the tracking target of track failure has the specific projection image of overlapping region；

Utilize progress target inspection in the specific projection image of the tracking target of tracking failure in (i+2) frame panoramic picture Survey, obtain testing result corresponding to the tracking target of tracking failure；

Obtained corresponding to tracking targets from all tracking failures in testing result testing result characterize detection successfully each with Track target, at least one tracking target that tracking target obtains with tracking successful each will be detected as (i+2) frame panorama All tracking targets in N number of projected image in image.
A kind of 13. panorama camera, it is characterised in that including readable storage medium storing program for executing, had program stored therein on the readable storage medium storing program for executing, Following steps are realized when described program is executed by processor：

Certain the frame panoramic picture chosen in panoramic video is initial frame panoramic picture, from the initial frame panoramic picture and afterwards N number of projected image is chosen in per frame panoramic picture, wherein, each projected image is the parts of images in corresponding panoramic picture Projection, N are the integer not less than 2；

M tracking target is chosen in the N number of projected image chosen from the initial frame panoramic picture, wherein M is whole not less than 1 Number；

Using the initial frame panoramic picture as the 1st frame panoramic picture, successively for every frame panorama sketch after the 1st frame panoramic picture As performing following steps, wherein, i gets s from 1 successively, and s is the integer more than 1：

It is tracked using each tracking target in N number of projected image in the i-th frame panoramic picture, it is complete obtains (i+1) frame Successfully at least one tracking target is tracked in N number of projected image in scape image.
14. panorama camera as claimed in claim 13, it is characterised in that described program is executed by processor described in Shi Cong just After M tracking target is chosen in N number of projected image that beginning frame panoramic picture is chosen, following steps are also realized：

Using the initial frame panoramic picture as starting point, at interval of N number of projection corresponding to k frames panoramic picture from respective frame panoramic picture At least one tracking target is chosen in image, wherein, k is the integer not less than 2；

Within the cycle of each interval k frame panoramic pictures, will choose it is at least one tracking target respective frame panoramic picture as 1st frame panoramic picture, following steps are performed for every frame panoramic picture after the 1st frame panoramic picture successively, wherein, i is successively K+1 is got from 1；

It is tracked using each tracking target in N number of projected image in the i-th frame panoramic picture, it is complete obtains (i+1) frame Successfully at least one tracking target is tracked in N number of projected image in scape image.
15. the panorama camera as described in claim 13 or 14, it is characterised in that is obtained when described program is executed by processor (i+1) when the successfully step of at least one tracking target is tracked in N number of projected image in frame panoramic picture, also realize following Step：

Whether each tracking target in N number of projected image in detection (i+1) frame panoramic picture tracks successfully；

In a tracking target following failure in often detecting (i+1) frame panoramic picture, according between projected image Syntople, obtain the neighboring projection images with the projected image adjoining where the tracking target of tracking failure；

The tracking target to be failed using tracking carries out target detection in the neighboring projection images in (i+2) frame panoramic picture, Obtain testing result corresponding to the tracking target of tracking failure；

Obtained corresponding to tracking targets from all tracking failures in testing result testing result characterize detection successfully each with Track target, at least one tracking target that tracking target obtains with tracking successful each will be detected as (i+2) frame panorama All tracking targets in N number of projected image in image.