Information based multimodal Background Subtraction for Traffic Monitoring Applications
Author
Alldieck, Thiemo Andreas
Term
4. term
Publication year
2015
Submitted on
2015-06-01
Pages
102
Abstract
Dette speciale præsenterer en ny metode til baggrundssubtraktion i multimodale systemer. Arbejdet bygger videre på Stauffer og Grimsons baggrundsmodel baseret på Gaussiske blandinger. Vi fusionerer værdier for baggrundsoverensstemmelse fra to billedkilder—termiske og RGB—så systemet mere stabilt kan skelne mellem baggrund og bevægelige objekter ved vedvarende overvågning. Vi opstiller heuristikker for billedkvalitet, baseret på billedegenskaber og eksterne kilder, til at vurdere nytten af hver modalitet og udføre kontekstbevidst fusion. Derudover præsenteres udvidelser målrettet trafikovervågning, herunder tilpasninger af en ny billedrepræsentation, der beskriver hvor godt hver pixel passer til baggrundsmodellen. Metodens potentiale er demonstreret gennem omfattende kvantitative og kvalitative tests.
This thesis introduces a new approach to background subtraction in multimodal systems that combine thermal and RGB cameras. The work extends the Gaussian mixture model background subtraction method by Stauffer and Grimson. We fuse background conformity values from the two image sources—that is, how well each pixel matches the background model—to achieve more stable performance in persistent surveillance. We specify image quality heuristics, based on image characteristics and external sources, to assess the usefulness of each modality and perform context-aware fusion. We also present extensions for traffic monitoring, including adjustments to a new image representation that encodes pixel conformity to the background model. The potential of the method is demonstrated through extensive quantitative and qualitative tests.
[This abstract was generated with the help of AI]
Keywords
Documents