1 / 13

CV Workshop: Multiple Target Tracking

CV Workshop: Multiple Target Tracking. Michael Rubinstein IDC Jan. 27 2009. Target Tracking and MTT. The problem: Identifying moving objects Practically: Input: Detection/Sensor (noisy) measurements Estimating the most probable measurement at time k from measurements up to time k

svein
Télécharger la présentation

CV Workshop: Multiple Target Tracking

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. CV Workshop:Multiple Target Tracking Michael Rubinstein IDC Jan. 27 2009

  2. Target Tracking and MTT • The problem: • Identifying moving objects • Practically: • Input: Detection/Sensor (noisy) measurements • Estimating the most probable measurement at time k from measurements up to time k • Applications: • Computer vision (tracking), robotics, control theory, astronomy, ballistics (missiles), econometrics (stocks), etc…

  3. MTT in Dense Crowd • Detection of head tops (+ height) using multiple cameras • Current method • Heuristic, but works well • Offline • In this work: • Mathematical model • Online Eshel & Moses, 2008

  4. The Kalman Filter • Assumptions: • The process is modeled by a linear system. e.g. xk=xk-1+vt • Measurement (and prediction) noise is normally distributed • Result: • Analytic solution! • Unique “best estimate”

  5. The Kalman Filter • Predictor(a-priori)-corrector(a-posteriori) model

  6. Tracking Multiple Targets

  7. Tracking Engine Detections classifier Update Targets Predict Targets

  8. Classifier T1 T2 T5 Y T3 T4 X

  9. Results

  10. Results

  11. Results

  12. Until now • What have I learned about this problem? • It’s a problem… • Many parameters, should be set as accurately as possible • Need labeled data • Pros • Sound model • Linear system + normal estimation might be sufficient • Not much references for dense tracking

  13. Future • Tuning! • maybe learn parameters from data • Will it do better than current method? • Combine shorter, higher-accuracy tracks • Particle Filter

More Related