Institute of Automation Chinsese Academy of Sciences
대리인 / 주소
Howard, Jeremy
인용정보
피인용 횟수 :
0인용 특허 :
8
초록▼
The present invention relates to a large-range-first cross-camera visual target re-identification method. The method comprises: step S1, obtaining initial single-camera tracks of targets; step S2, calculating a piecewise major color spectrum histogram feature of each track, and obtaining a track fea
The present invention relates to a large-range-first cross-camera visual target re-identification method. The method comprises: step S1, obtaining initial single-camera tracks of targets; step S2, calculating a piecewise major color spectrum histogram feature of each track, and obtaining a track feature representation; step S3, obtaining a calculation formula of the similarity between any two tracks by using a minimum uncertainty method, so as to obtain the similarity between any two tracks; and step S4, performing global data association on all the tracks by using a maximum posterior probability method, so as to obtain a cross-camera tracking result. The target re-identification method of the present invention achieves high correct identification accuracy.
대표청구항▼
1. A large-range-first cross-camera visual target re-identification method, characterized in that said method comprises: step S1: obtaining initial single-camera tracks of targets;step S2: calculating a piecewise major color spectrum histogram feature of each track and obtaining a track feature repr
1. A large-range-first cross-camera visual target re-identification method, characterized in that said method comprises: step S1: obtaining initial single-camera tracks of targets;step S2: calculating a piecewise major color spectrum histogram feature of each track and obtaining a track feature representation;step S3: obtaining a calculation formula for the similarity between any two tracks by using a minimum uncertainty method so as to obtain the similarity between any two tracks; andstep S4: performing global data association on all the tracks by using a maximum posterior probability method to obtain a cross-camera tracking result. 2. The method according to claim 1, characterized in that in said step S1, for each track, a mean value of confidence of all frames is used to represent a track accuracy of the track: From here it is clear the claim is calculating specific characteristics of data and then utilizing specific methodology on particular characteristics of the tracks: c=∑j=tsteαj/(te-ts)(1)wherein the confidence α represents the result of tracking of each frame, α<0.2 means that the tracked target is lost, and ts and te are respectively the start frame and end frame of the track;a finally formed set of tracks of all targets is L={l1, l2, . . . , lN}, wherein N is a track summary, and each track li=[xi, ci,si,ti,ai] represents the position, accuracy, scene, time and apparent features of the track, respectively. 3. A large-range-first cross-camera visual target re-identification method, the method comprising: step S1: obtaining initial single-camera tracks of targets;step S2: calculating a piecewise major color spectrum histogram feature of each track and obtaining a track feature representation;step S3: obtaining a calculation formula for the similarity between any two tracks by using a minimum uncertainty method so as to obtain the similarity between any two tracks; andstep S4: performing global data association on all the tracks by using a maximum posterior probability method to obtain a cross-camera tracking result,wherein step S2 includes calculating color histograms of targets of each frame, then dividing the color space into 16*2 colors according to the values of H and S, and selecting the first n color values as the features of said targets in said frame: h={C1,C2, . . . ,Cn} (2) wherein Ci is one color of the first n colors whose sum of the pixel numbers accounts for above 90% of that of the total pixel numbers, and a general feature of each track is: H=Σi=1mkhi (3) wherein mk is the length of track k; calculating similarities therebetween as Λ=Sim(hi,hj) for all features hi in the general feature H, and finding a movement period through information of similarities between each frame in the track, then re-segmenting the original track feature H according to the period, wherein the periodic information p that might exist in the general feature H is obtained by: p=argmaxt1mk-t∑j=1mk-tΛj,j+t(4) and the track is re-segmented uniformly according to the periodic information p so as to obtain a piecewise major color spectrum histogram feature of the track: H={H1,H2, . . . ,Hd} (5) in which d=┌mk/p┐ represents the number of segments into which the track is segmented. 4. The method according to claim 3, characterized in that said step S3 specifically includes: calculating a similarity between two tracks to guide matching between the tracks, and maximizing the similarity while minimizing the uncertainty, so that the obtained similarity match value can reflect the real similarity relation between two tracks, wherein the matching formula is: Dis(HA,HB)=1-maxSim(HiA,HjB)-minSim(HuA,HvB)maxSim(HiA,HjB)+minSim(HuA,HvB)(6) in which HA and HB are piecewise major color spectrum histogram features of two tracks, and HiA and HjB are certain segments thereof, i={1, 2, . . . , dA}, j={1, 2, . . . , dB}. 5. The method according to claim 4, characterized in that said step S4 specifically includes: step S4-1: obtaining each globally associated track T={li1, li2, . . . lik}, and obtaining a general set of associated tracks T={T1, T2, . . . , Tm}, m being the number of associated tracks; then obtaining a maximum posterior probability of set T when a given set L of tracks and the associated tracks do not overlap: T*=argmaxT∏iP(li❘T)∏Tk∈TP(Tk)Ti⋂Tj=ϕ,∀i≠j(7) wherein P(li/T) is the similarity of track li, and P(Tk) is a possible priori probability of associated tracks, which can be represented by a Markov chain containing a transition probability ΠP(lki+1|lki); step S4-2: building a graph structure, wherein each node represents a track li and its value is ci, each edge represents a priori probability P(li→lj), and obtaining a set that enables T* to be the maximum from the minimum cost function flow of the entire graph, wherein the cost energy eij of each flow is represented by a negative logarithmic function as: eij=-logP(L❘li→lj)P(li→lj)=-log(Pm*Pt*Pa)(8) in which Pm and Pt respectively represent match probabilities of motion information and time information between tracks, and Pa represents the match probability of apparent features of tracks, whose matching similarity formula is: Pa={Dis(HA,HB)ifsi=sjλDis(HA,HB)ifsi≠sj(9) and the cost energy of each flow is obtained, then a traversing is performed to finally obtain a set T that enables the posterior probability to be the maximum, which is just the result of multi-camera target tracking and re-identification. 6. The method according to claim 1, wherein step S2 includes finding a movement period through information of similarities of an original track feature between each frame in the track, then re-segmenting the original track feature according to the period. 7. A non-transitory storage medium tangibly embodying a program of machine-readable instructions executable by a digital processing apparatus to perform a method, said method comprising: step S1: obtaining initial single-camera tracks of targets;step S2: calculating a piecewise major color spectrum histogram feature of each track and obtaining a track feature representation;step S3: obtaining a calculation formula for the similarity between any two tracks by using a minimum uncertainty method so as to obtain the similarity between any two tracks; andstep S4: performing global data association on all the tracks by using a maximum posterior probability method to obtain a cross-camera tracking result. 8. The non-transitory storage medium according to claim 7, characterized in that in said step S1, for each track, a mean value of confidence of all frames is used to represent a track accuracy of the track: From here it is clear the claim is calculating specific characteristics of data and then utilizing specific methodology on particular characteristics of the tracks: c=∑j=tsteαj/(te-ts)(1) wherein the confidence α represents the result of tracking of each frame, α<0.2 means that the tracked target is lost, and ts and te are respectively the start frame and end frame of the track; a finally formed set of tracks of all targets is L={l1, l2, . . . , lN}, wherein N is a track summary, and each track li=[xi,ci,si,ti, ai] represents the position, accuracy, scene, time and apparent features of the track, respectively. 9. The non-transitory storage medium according to claim 8, characterized in that said step S2 specifically includes: calculating color histograms of targets of each frame, then dividing the color space into 16*2 colors according to the values of H and S, and selecting the first n color values as the features of said targets in said frame: h={C1,C2, . . . ,Cn} (2) wherein Ci is one color of the first n colors whose sum of the pixel numbers accounts for above 90% of that of the total pixel numbers, and a general feature of each track is: H=Σi=1mkhi (3) wherein mk is the length of track k; calculating similarities therebetween as Λ=Sim(hi,hj) for all features hi in the general feature H, and finding a movement period through information of similarities between each frame in the track, then re-segmenting the original track feature H according to the period, wherein the periodic information p that might exist in the general feature H is obtained by: p=argmaxt1mk-t∑j=1mk-tΛj,j+t(4) and the track is re-segmented uniformly according to the periodic information p so as to obtain a piecewise major color spectrum histogram feature of the track: H={H1,H2, . . . ,Hd} (5) in which d=┌mk/p┐ represents the number of segments into which the track is segmented. 10. The non-transitory storage medium according to claim 9, characterized in that said step S3 specifically includes: calculating a similarity between two tracks to guide matching between the tracks, and maximizing the similarity while minimizing the uncertainty, so that the obtained similarity match value can reflect the real similarity relation between two tracks, wherein the matching formula is: Dis(HA,HB)=1-maxSim(HiA,HjB)-minSim(HuA,HvB)maxSim(HiA,HjB)+minSim(HuA,HvB)(6) in which HA and HB are piecewise major color spectrum histogram features of two tracks, and HiA and HjB are certain segments thereof, i={1, 2, . . . , dA}, j={1, 2, . . . , dB}. 11. The non-transitory storage medium according to claim 10, characterized in that said step S4 specifically includes: step S4-1: obtaining each globally associated track Ti={li1, li2, . . . , lik}, and obtaining a general set of associated tracks T={T1, T2, . . . , Tm}, m being the number of associated tracks; then obtaining a maximum posterior probability of set T when a given set L of tracks and the associated tracks do not overlap: T*=argmaxT∏iP(li❘T)∏Tk∈TP(Tk)Ti⋂Tj=ϕ,∀i≠j(7) wherein P(l1|T) is the similarity of track li, and P(Tk) is a possible priori probability of associated tracks, which can be represented by a Markov chain containing a transition probability √P(lki+1|lki); step S4-2: building a graph structure, wherein each node represents a track li and its value is ci, each edge represents a priori probability P(li→lj), and obtaining a set that enables T* to be the maximum from the minimum cost function flow of the entire graph, wherein the cost energy eij of each flow is represented by a negative logarithmic function as: eij=-logP(L❘li→lj)P(li→lj)=-log(Pm*Pt*Pa)(8) in which Pm and Pt respectively represent match probabilities of motion information and time information between tracks, and Pa represents the match probability of apparent features of tracks, whose matching similarity formula is: Pa={Dis(HA,HB)ifsi=sjλDis(HA,HB)ifsi≠sj(9) and the cost energy of each flow is obtained, then a traversing is performed to finally obtain a set T that enables the posterior probability to be the maximum, which is just the result of multi-camera target tracking and re-identification.
연구과제 타임라인
LOADING...
LOADING...
LOADING...
LOADING...
LOADING...
이 특허에 인용된 특허 (8)
Lv, Fengjun; Xu, Wei; Gong, Yihong, Efficient multi-hypothesis multi-human 3D tracking in crowded scenes.
Zhao, Tao; Aggarwal, Manoj; Kumar, Rakesh; Sawhney, Harpreet, Method and apparatus for tracking objects over a wide area using a network of stereo sensors.
Center, Jr.,Julian L.; Wren,Christopher R.; Basu,Sumit; Gusyatin,Evgeniy, Methods of establishing a communications link using perceptual sensing of a user's presence.
Lipton, Alan J.; Strat, Thomas M.; Venetianer, Pèter L.; Allmen, Mark C.; Severson, William E.; Haering, Niels; Chosak, Andrew J.; Zhang, Zhong; Frazier, Matthew F.; Seekas, James S.; Hirata, Tasuki; Clark, John, Video surveillance system employing video primitives.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.