1 / 13

Motion Estimation of Moving Foreground Objects

Motion Estimation of Moving Foreground Objects. Pierre Ponce ee392j Winter 2003-04. March 10, 2004. Outline. Motivation System Overview Models Used Object Segmentation Trajectory Projection Conclusion. Motivation.

matilde
Télécharger la présentation

Motion Estimation of Moving Foreground Objects

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. Motion Estimation of Moving Foreground Objects Pierre Ponceee392jWinter 2003-04 March 10, 2004

  2. Outline • Motivation • System Overview • Models Used • Object Segmentation • Trajectory Projection • Conclusion

  3. Motivation • Making the foundation for a system that isolates the objects we are interested in tracking. • Take the background out of consideration when estimating motion. • Interest rests only on the layer(s) inmotion.

  4. Is this practical? Yes, and it has many uses. For example, • Traffic (Intersections, Highways) • Sports Events (many cameras) • Surveillance (fixed cameras)

  5. System View Images BG Modeling FG Isolation Median Filtering Motion Rep. Motion Gradient Estimation Centroid Detection Object “Blobbing”

  6. What is “background”? • A stationary layer located behind all other layers. • Undergoes apparent motion (wind, lighting changes, shadowing, etc.) • These assumptions can be exploited to form a stochastic model (per pixel) of the behavior of the image.

  7. BG Modeling Each pixel in an image is composed of background unless it is occluded by a foreground object. Model: Each pixel is the sum of Gaussian processes with the image color information as its variable.

  8. Gaussian Mixture at time t: Multiple Gaussians are necessary to compensate for different types of interactions between the background pixels and scene factors (lighting, shadows, object interactions, etc.)

  9. k-Means Clustering • The parameters for the k Gaussian components of each pixel are computed through this iterative technique. • Updates occur on each frame. • Each frame contributes information about the background characteristics. Background images tend to have lower variance than moving objects. If a moving object becomes static over a long period of time, it will eventually become part of the background (how long depends of the characteristics of the object)

  10. Foreground Isolation • Anything that is not background is likely foreground. • For each pixel, the distributions that are more likely and have less variance are usually part of the background. From: Stauffer, C and W.E.L. Grimson. Adaptive background mixture models for real-time tracking. [Apr. 1999]

  11. Object Segmentation • The foreground layer is median-filtered to compensate for artifacts from the BG estimation. • The individual foreground objects should be considered to be dense “blobs”, which can be identified by their centroid.

  12. Motion Trajectory The direction of motion for each blob object can be estimated by only dealing with motion vectors computed from within foreground sections. This reduces the amount of computation necessary to extract motion information. (useful in coding where layers are used)

  13. Conclusion • Background Modeling is very robust to different noise factors and seems to be a reliable stepping stone for automatic layering. (Drawback: can be slow) • Object Segmentation becomes easier with a reliable background model. • Motion trajectories can be computed with any of the known parametric models for motion estimation.

More Related