1 / 29

Toward Managing Uncertain Spatial Information for Situational Awareness Applications

Toward Managing Uncertain Spatial Information for Situational Awareness Applications. Authors: Yiming Ma, Dmitri V. Kalashnikov and Sharad Mehrotra Presented by: Jonathan Durda. Situational Awareness Applications.

bella
Télécharger la présentation

Toward Managing Uncertain Spatial Information for Situational Awareness Applications

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Toward Managing Uncertain Spatial Information for Situational Awareness Applications Authors: Yiming Ma, Dmitri V. Kalashnikov and SharadMehrotra Presented by: Jonathan Durda

  2. Situational Awareness Applications • Take different inputs and aid in decision making, reasoning and analysis of emergency situations • Example: September 11, 2001 • 1. “... The PAPD Mobile Command Post was located on West St. north of WTC and there was equipment being staged there. .. • 2. “... a PAPD Command Truck parked on the west side of Broadway St. and north of Vesey St. ...”

  3. Situational Awareness Applications • Algorithms take this data as input and attempt to reason where the event is taking place • Information can be used to help emergency personnel and first responders to a disaster to have a better idea of where the event is taking place

  4. Introduction • Uncertain locations are represented as random variables • Algorithm assumes that when events are reported, those reporting use landmarks to describe location • In our example, the landmark would be the World Trade Centers • Spatial descriptors refer to words that describe location, including “near WTC, in front of WTC, behind WTC”

  5. Situational Awareness • Probabilistic Model used for uncertain spatial information • Formality – this model adds to the probability theory • Practicality – it has been implemented • Generality – the model can handle many types of queries about different spatial information • Effectiveness – other models and solutions that are already in place are known to work

  6. Situational Awareness • A grid is used representing all of the information of possible space • Using input, the model gives the probability that an event happened in a given cell • Histograms used to represent locations, data is stored in a database in quad-tree form • Uncertain locations are indexed using R-trees

  7. Situational Awareness • Each node in R-tree stores additional information about the left and right “x bound” of the uncertain event • Also stores top and bottom “y bounds” for location • A rectangle is formed based on the x bounds • Range queries include indoor and outdoor events

  8. Modeling Location Uncertainty • The goal when a request is made is to give the location of the event by f(x,y|report) • F(x,y,|s,t) • “A traffic accident near World Trade Center” • S would be “near(WTC)” and t would be “traffic accident” • S gives the best location of the event • T helps us determine where the event may have happened, in this case in the road

  9. Modeling Location Uncertainty • Four classes of s-descriptors: topoligical relations, cardinal direction relations, orientation relations and distance relations • S-expression contains many s-descriptors • “I am near building A and near building B” • This would create the s-expression {near(A) , near (B)}

  10. Modeling Location Uncertainty • Pdf is Probability Density Function

  11. Modeling Location Uncertainty • An s-descriptor can be near a building • Outside of a building is also s-descriptor • If an event is not inside of a building, the probability density function (pdf) is zero for inside the building • Bayes formula: f(x,y|s1,s2) = P(s1,s2|x,y) f(x,y) • P(s1,s2) • P(s1,s2) is the probability of observing events s1 and s2 given location x and y

  12. Modeling Location Uncertainty • Events s1 and s2 are independent, therefore: • F(x,y|s1,s2) = f(x,y|s1) x f(x,y|s2) • Event t can be viewed uniform or non uniform • F(x,y|t) tells where a given even is likely to occur • If t is “home robbery”, the event occurred in a home and not in the street

  13. Spatial Queries • A range query is defined to be all elements whose probability of being inside R is greater than zero • Probabilistic query is detached if it returns elements without probabilities assigned to them • The query is attached if it returns a set of tuples with elements and probabilities assigned to them • By default, spatial queries are detached

  14. Spatial Queries • A query is said to have threshold semantics if a query Q returns all elements that have probabilities greater than threshold p • We represent shapes and objects in a grid with notation G(I,j) with i and j being coordinates in the grid • A region R must occupy either all of a cell or no part of a cell at all • vcellsin(R) denotes how many cells are in a range R

  15. Efficient pdf Representation • Histograms represent pdfs, but may take up a large amount of space • Amount of space needed can be reduced by determining that some locations do not need to be in as much detail as others • Query processing can be more efficient by keeping only certain information about locations • Histogram is indexed using a quad-tree • Can be used to index the pixels of an image

  16. Efficient pdf Representation

  17. U-Grid • Uncertain grid (U-grid) is a 2 dimensional array of cells

  18. U-Grid • Each cell Lij also has pmax and psum attributes that has the maximum pmax and psum value of the entire list • Processing of range queries has indexing and object phases • Quad tree is traversed to find probability of query • Information stored in the grid is used to find a good upper bound for a location

  19. U-Grid • Cell-level pruning is done by finding the minimum of the number of cells in a range x pmax,psum • If cell level pruning does now work, the list Lij is processed sequentially by location • Location pruning is the lowest level of Pruning. If it cannot be pruned, it is inserted into a processor list

  20. Multiple Lists • Having multiple lists per cell increases efficiency and each list has a threshold • When algorithm goes to process a list of lists, it starts with the rightmost list • Each list looked at for pruning • Lists are created to maximize the pruning power of a range query

  21. Multiple Lists • Minimize size of bounding regions (BR) so that this region does not intersect with a query region • BR should be chosen to minimize I/O time and increase efficiency • BR should be selected so that storing and indexing locations is very fast • Primary list of a cell is the list itself, and all other lists are secondary lists • Bounding regions are split into four corners so that algorithm runs efficiently

  22. Index Slicing • Store many x bounds to increase pruning power • Increases size of R tree however • So we use multiple indexes with information in each one • Create kxbounds for each threshold level • Algorithm starts with rightmost R tree, this technique does not greatly increase size of R tree

  23. Processing Range Queries with U-Grids • Computing exact probabilities for each cell can be very time consuming • We find a good upper bound using information in the directory grid

  24. Evaluating this Strategy • Authors experimented on 2 GHz PC with 1 GB RAM • Data used is from 164 NYPD reports on September 22, 2001 • The retrieval time for each model is examined

  25. Experimenting

  26. Experimenting • U-Grid is the best solution for a city level area • Events separated into low (less than 10x10 vcells) medium (less than 20x20) and high (less than 100x100) • Data uncertainty controlled by including many objects with different uncertainty levels

  27. Experimenting

  28. Experimenting

  29. Experimenting

More Related