Tracking of objects with known color signature - ELITECH 20
Download as PPT, PDF0 likes798 views
The document discusses an approach for tracking objects using their color signatures, emphasizing preprocessing, image segmentation, object identification, and tracking methods. It outlines existing techniques, presents a specific methodology involving histogramization and convex hulls, and highlights improvements in tracking speed and robustness. Future work includes developing adaptive representations and training algorithms to enhance computational efficiency.
![Tracking of objects with known color signature Lukáš Tencer [email_address]](https://image.slidesharecdn.com/prezent-111207174844-phpapp02/85/Tracking-of-objects-with-known-color-signature-ELITECH-20-1-320.jpg){kind=tracking}
![Previous work ACM Survey [Yilmaz et al., 2004], after this emphasis on accuracy, speed, robustness Kernel based tracking tracking kernel of image region Point Based Tracking](https://image.slidesharecdn.com/prezent-111207174844-phpapp02/85/Tracking-of-objects-with-known-color-signature-ELITECH-20-3-320.jpg){kind=tracking}
Tracking of objects with known color signature - ELITECH 20
- 1. Tracking of objects with known color signature Lukáš Tencer [email_address]
- 2. Content Overview of existing approaches Motivation Basic apparatus Our approach: Preprocessing Image segmentation Object identification Object tracking Testing and Results Future work
- 3. Previous work ACM Survey [Yilmaz et al., 2004], after this emphasis on accuracy, speed, robustness Kernel based tracking tracking kernel of image region Point Based Tracking
- 4. Previous work 2 Structure Based Tracking uses skeleton/silhouette Our previous work Gesture-based mouse control
- 5. Basic apparatus Color signature: artificial representation of objects projection into digital space, based on its color values for our method, it is ranged values in HSV color model Convex hull of set of points: Polygon, for which every point from input set lies “inside” the polygon minimal convex set containing input set
- 6. Our approach Preprocessing Background subtraction Image Segmentation Image histogramization Object Identification Color contribution condition, border conditions Object Tracking Bumper-region based tracking
- 7. Preprocessing Background subtraction Color difference between trained background and input image Could cause loss of image information
- 8. Image segmantation Image Histogramization Segment image into sub-regions (fixed/adaptive) Extract descriptive channel of the image In our case, HUE channel of HSV image model Create histogram with n dimensions for each sub-region
- 9. Object Identification Select regions corresponding to object’s projection into image plane For one colored object, we select minimal level of dominant dimension For two colored object, minimal level of contribution of other color Border regions, maximal contribution of other colors and minimal connectivity condition
- 10. Object tracking Pick center of the convex hull of identified regions Place tracking object inside the convex hull Once object is moved, we identify regions, which are no longer covered by object Create convex hull of uncovered regions, connect it’s center and center of tracking object to identify movement vector Move tracking object in direction of movement vector, until all regions are covered by object regions
- 11. Testing and Results Improved tracking speed Robustness against disruptive elements in scene Unique color signature Approach of image histogramization Bumper-region based object tracking Method could be used for tracking of colored object, HCI, surveillance …
- 12. Future Work Use of more adaptive object representation structures, to improve computational speed Quad-tree? Remove requirement to have prior information about color signature Use training algorithm, to create model from identified foreground
- 13. Thank you for you attention
- 14. Questions ?