130 likes | 147 Views
This paper presents an efficient method, STMPM algorithm, for mining spatio-temporal moving patterns from large sets of spatial-temporal moving data. The algorithm efficiently extracts periodic or sequential frequent moving patterns by creating a moving sequence tree, reducing the overall time required for pattern mining.
E N D
Efficient STMPM (Spatio-Temporal Moving Pattern Mining) Using Moving Sequence Tree Reporter :Wen-ho Tsai Fourth International Conference on Networked Computing and Advanced Information Management , IEEE 2008 Page 432-437 YonSIk Lee , Hyun Ko Dept. of Computer Information Engineering, Kunsan National University, Kunsan, Chonbuk, 573-701, Korea
Outline • Introduction • Preprocessing • Moving Sequence Tree • STMPM algorithm • Experiment • Conclusion
Introduction • To solve the size of memory required for search greatly increases if a set of candidate moving sequences or data is large, STMPM algorithm is proposed as the new method to efficiently extract the periodical or sequential frequent moving patterns from enormous sets of spatial-temporal moving data.
Preprocessing • [Definition]Given moving object database MD, user-specified minimum support factor min_sup, and the constraint of time interval between spatial scopes max_gap, spatio-temporal frequent moving pattern mining searches all frequent moving sequences that satisfies the minimum support factor.
Preprocessing Symbol Table Location Number
Preprocessing • If <Start location>=<End location>, Set this symbol=<N>. • For instance, passing ball will have eight symbol moving sequences :
Moving Sequence Tree • Sequence trees of the moving sequence of Passing ball. 3 L N D F 1 R F D F 2 4 L N D F 2 R F D F 1
STMPM Algorithm • Input : D(Database), Cs(Spatial Constraint), Ct(Temporal Constraint), • max_gap(Time Interval Constraint), Lsk(Spatial Level), • Ltk(Temporal Level), min_sup(Threshold of Minimum Support) • Output : FrePatternSet • Procedure STMP(D, Cs, Ct, max_gap, Lsk, Ltk, min_sup) • Begin • PrevA = null; • // Set of History Data staisfied Constraint of Spatial and Temporal • Drt = LimitedDataSet (D, Cs, Ct); • // Extracting Set of Moving Sequence from Drt • Dseq = SeqExtractor (Drt, max_gap); • // Generalizing Region to Spatial Property of Moving Sequence in Dseq • DRgen = Contains (Dseq, PrevA, Lsk); • // Generalizing Valid Time to Temporal Property of Moving Sequence in DRgen • DTgen = During (DRgen, Ltk); • // Frequent Moving Pattern Mining • FreqPatternSet = FreqPatternExtractor (DTgen, min_sup); • Return FreqPatternSet; • End
Frequent moving sequences • My sport video frequent moving sequences in passing ball.(data set size=50,min_sup=10%) 3 L N D F 1:20 4 L N D F 2:15
Conclusion • The mining processing times for each minimum support factor were compared among the proposed algorithm, STMPM, MPMine and Apriori-MSP as the number of moving object changes. The result of the experiments showed that STMPM algorithm is much more efficient than the other algorithms, and this is because it searches patterns by creating moving sequence tree which minimizes the number of accesses to database, thus, reducing overall time required for pattern mining.
Conclusion • Although execute time is faster than template matching method, it’s precision down to 56%