1 / 136

TRACKING Particle Filter

TRACKING Particle Filter. Computer Vision Seminar – Hagit Hel-Or Itzik Berrebi Rotem Klorin. Tracking. Tracking - the process of locating a moving object (or multiple objects) over time using a camera .

edwinl
Download Presentation

TRACKING Particle Filter

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. TRACKING Particle Filter Computer Vision Seminar – Hagit Hel-Or ItzikBerrebi RotemKlorin

  2. Tracking • Tracking - the process of locating a moving object (or multiple objects) over time using a camera. • Variety of uses: human-computer interaction, security and surveillance, video communication, traffic control…

  3. Tracking - Goal • Finding the objects. • Giving them an identity. • Follow the object as they move around while maintaining their identity – even if the objects collide, disappear for a short time, etc.

  4. Particle Filter

  5. Main differences with Kalman filter: Kalmanfilter: • Assumes uni-modal (Gaussian) distribution. • Predicts single new state for each object tracked. • Updates state based on error between predicted state and observed data.

  6. Main differences with Kalman filter: Particle filter: • Can work for multi-modal distribution. • Predicts multiple possible states for each object tracked. • Each possible state has a different probability. • Estimates probabilities of predicted states based on observed data.

  7. Particle Vs Kalman

  8. Particle Filter • Particle filter or Sequential Monte Carlo (SMC) methods are a set of algorithms that estimate the posterior density of the state variables given the observation variables.

  9. Particle Filter Predicts multiple possible next states: • Achieved using sampling or drawing samples from the probability density functions.

  10. Recursive Filter • The particle filter is a Bayesian sequential importance sampling technique, which recursively approximates the posterior distribution using a finite set of weighted samples. • Each state is estimated by the information from the last state.

  11. Players • Observations and hidden states. • Particles and weights. • Probability

  12. Observations and Hidden States • The particle filter is designed for a Hidden Markov Model, where the system consists of hidden and observable variables.

  13. Observations and Hidden StatesExample - hand tracking Observations: • Images (R,G,B matrices)

  14. Observations: • Binary contour images (0/1 matrices)

  15. Observations: • Edge detection images

  16. Observations: • Harris corners (lists of all detected Harris corners) • Other visual features collected from an image

  17. Hidden States: • Hand contour images

  18. Hidden States: • List of splines

  19. Hidden States: • List of phalanx • List of angles between phalanxes

  20. Hidden States: • One of the following: { paper, rock, scissors, lizard, spock}

  21. Assumptions • - Hidden states • - Observations • Each depends only on

  22. Assumptions • - Hidden states • - Observations • Each depends only on • is a Markov Chain:

  23. Example Robot plays rock-paper-scissors DEMO

  24. Players • Observations and hidden states. • Particles and weights. • Probability

  25. Particles and Weights • Particles - Samples of hypotheses. • The hypotheses are estimated positions of the object we are tracking. • The more particles we sample, the faster we find the object.

  26. Particles and Weights • Each particle is given a weight according to its probability of being the next state. • Particles with a bigger chance to be the next state will get higher weights.

  27. Particle and Weights • High probability samples should be drawn more frequently. • Low probability samples should be drawn less frequently.

  28. Particle Filter - Example DEMO

  29. Players • Observations and hidden states. • Particles and weights. • Probability

  30. HMM – Hidden Markov Models • P(X=H | O=T) = ? • P(X)= • Observation Model: P(O|X)= • Transition Model:

  31. HMM – Hidden Markov Models

  32. HMM – Hidden Markov Models • Bayes’ update: • Prediction:

  33. Particle Filter • Posterior: What is the probability that the object is at the location for all possible locations if the history of observations is ? • Prior: The motion model – where will the object be at time instant t given that it was previously at ? • Likelihood: The likelihood of making the observation given that the object is at the location

  34. Goal – Posterior Density • We would like to calculate: • Given the data:

  35. Particle Filter Algorithm

  36. Particle Filter - Algorithm • Sample the particles using a model. • Compute weights according to the model. • Resampling with respect to the weights: replace unlikely samples by more likely ones.

  37. Particle Filter - Example DEMO

  38. Aircraft example -Assumptions: • can be estimated from • can be calculated from • Velocity = 800 to 1000 km/h

  39. In each iterator: • Sampling - picking big set of possible plane locations. • Prediction - assigning weights using a map: | • Update - updating locations and weights. |

  40. A weighted set of N particles: P={(, )|1≤i≤N} In each iteration: • create new set P with respect to the weights in old P. • set the weight of each particle in new P. • update the locations and weights of particles in P using the model.

  41. Particle Filter • Particle filterconsists of essentially two steps: Prediction and update.

  42. Prediction • Given all available observations: up to time t-1, the prediction stage uses the probabilistic system transition model: to predict the posterior at time t as:

  43. Update At time t, the observation is available, the state can be updated using Bayes’ rule: Where is described by the observation equation.

  44. Particles and weights The posterior: is approximated by a finite set of N samples with importance weights .

  45. Particles and weights The candidate samples are drawn from an importance distribution = And the weight of the samples are: =

  46. Numerical Example Tracking a robot: • We attempt to track the position of a robot across two states using four particles. • States: A,B (=A means that particle i is at position A at time t)

  47. Numerical Example Step 1 – model: • , The robot starts out as equally likely to be in either state.

  48. Numerical Example • The motion model: , Any object has probability of moving to or staying in B of and any object of moving to or staying in A of .

  49. Numerical Example • The observation model: If an objects is in state s in {A,B}, the probability that our observation is accurate is 80%.

More Related