A Region-based Object Tracking Method Using AdaBoost-based Feature Selection
碩士 === 國立中正大學 === 資訊工程所 === 94 === We proposed an integrated tracking system for applications which need more precise segmentation of target. The main tracking mechanism is accomplished by two trackers. The first tracker performs tracking by Adaboost on pixel-based seed features; it can provide more...
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Others |
| Language: | en_US |
| Published: |
2007
|
| Online Access: | http://ndltd.ncl.edu.tw/handle/91912251890535846637 |
| Summary: | 碩士 === 國立中正大學 === 資訊工程所 === 94 === We proposed an integrated tracking system for applications which need more precise segmentation of target. The main tracking mechanism is accomplished by two trackers. The first tracker performs tracking by Adaboost on pixel-based seed features; it can provide more detailed segmentation of target. The second tracker achieves tracking by bidirectional k-means clustering on regions, and uses Adaboost on region-based seed features to provide compensations to the short the first tracker. We also implement a tool which allows users to refine the result manually. By confidence measure, users can easily choose the timing to interact with frames which probably are in occlusion or perform poorly due to some complex situation.
|
|---|