we consider situations in which the object is unknown

1
we consider situations in which the object is unknown the only way of doing pose estimation is then building a map between image measurements (features) and configurations (poses) directly from the data this can be done in a training stage, where an collected as values (motion capture) hidden Markov models (HMMs) provide a way to the learn feature-pose maps automatically through the EM algorithm • for each state of the HMM a Gaussian likelihood is set up on the feature range • this partitions the feature range into n regions (approximate feature space) • and is associated with a refining from Y to Q PERFORMANCES PERFORMANCES pose estimate for the leg exp EXPERIMENTS ON HUMAN BODY TRACKING EXPERIMENTS ON HUMAN BODY TRACKING FEATURE-POSE MAPS AND HMMs FEATURE-POSE MAPS AND HMMs FEATURE EXTRACTION FEATURE EXTRACTION two experiments: four markers on right arm, eight markers on legs • we built evidential models for the two separate views and compared them with an overall model EVIDENTIAL MODELING FOR POSE ESTIMATION EVIDENTIAL MODELING FOR POSE ESTIMATION Fabio Cuzzolin, Computer Science Department, UCLA; Ruggero Frezza, DEI, Fabio Cuzzolin, Computer Science Department, UCLA; Ruggero Frezza, DEI, Universita’ di Padova Universita’ di Padova 4 4 nd nd INTERNATIONAL SYMPOSIUM ON IMPRECISE PROBABILITIES AND THEIR APPLICATIONS, INTERNATIONAL SYMPOSIUM ON IMPRECISE PROBABILITIES AND THEIR APPLICATIONS, ISIPTA’05 Carnegie Mellon University, Pittsburgh, USA, July 20-23 2005 ISIPTA’05 Carnegie Mellon University, Pittsburgh, USA, July 20-23 2005 TRAINING A BOTTOM-UP MODEL TRAINING A BOTTOM-UP MODEL the collection of approximate parameter space and feature spaces, linked by the refining maps learned in the training stage form an evidential model of the object the evidential model can then be used thereafter to estimate the pose of the body when new images become available BOTTOM LINES BOTTOM LINES we want to estimate the we want to estimate the configuration configuration (pose) of an (pose) of an unknown unknown object from a sequence of object from a sequence of images images model-free pose estimation requires model-free pose estimation requires to to build maps from build maps from image measurements (features) to image measurements (features) to poses poses different features have to be different features have to be integrated integrated to increase to increase accuracy and robustness of the accuracy and robustness of the estimation estimation the the evidential model evidential model provides such a provides such a framework framework MODEL-FREE POSE ESTIMATION MODEL-FREE POSE ESTIMATION pose estimation: reconstruction of the actual pose of a moving object by processing the sequence of images taken during its motion model-based pose estimation: a kinematic model of the body is known and used to help the estimate of component 9 of the pose vector shows a neat improvement when using the comprehensive model the silhouette of the body of interest is detected by colorimetric analysis • the bounding box containing the object is found • feature vector = collection of the coordinates of the vertices of the box example: Rehg and Kanade • pose = angles between links of the fingers 1 2 3 TRAINING : the body moves in front of the camera(s), while a sequence of poses is provided by a motion capture system. Then some features are computed from the images these feature sequences are passed to an HMM with n states yielding: two views acquired through DV cameras 1 ESTIMATION : the body performs new movements in front of the camera(s), and for each available image the features are computed as before the likelihoods of the features are transformed into belief functions those measurement functions are projected onto Q and combined a pointwise estimate of the object pose is computed by pignistic or relative plausibility transformation 2 comparison between visual estimate and real T k k I k p I .. 1 ) ( ) ( ˆ ˆ

description

4 nd INTERNATIONAL SYMPOSIUM ON IMPRECISE PROBABILITIES AND THEIR APPLICATIONS, ISIPTA’05 Carnegie Mellon University, Pittsburgh, USA, July 20-23 2005. EVIDENTIAL MODELING FOR POSE ESTIMATION Fabio Cuzzolin, Computer Science Department, UCLA; Ruggero Frezza, DEI, Universita’ di Padova. - PowerPoint PPT Presentation

Transcript of we consider situations in which the object is unknown

Page 1: we consider situations in which  the object is      unknown

we consider situations in which the object is unknown

the only way of doing pose estimation is then building a map between image measurements (features) and configurations (poses) directly from the data

this can be done in a training stage, where an approximate parameter space is collected as ground truth configuration values produced by an “oracle” (motion capture)

• hidden Markov models (HMMs) provide a way to the learn feature-pose maps automatically through the EM algorithm• for each state of the HMM a Gaussian likelihood is set up on the feature range• this partitions the feature range into n regions (approximate feature space)• and is associated with a refining from Y to Q

PERFORMANCESPERFORMANCES

• pose estimate for the leg exp

EXPERIMENTS ON HUMAN BODY TRACKINGEXPERIMENTS ON HUMAN BODY TRACKING

FEATURE-POSE MAPS AND HMMsFEATURE-POSE MAPS AND HMMs

FEATURE EXTRACTIONFEATURE EXTRACTION

• two experiments: four markers on right arm, eight markers on legs• we built evidential models for the two separate views and compared them with an overall model

EVIDENTIAL MODELEVIDENTIAL MODEL

EVIDENTIAL MODELING FOR POSE ESTIMATIONEVIDENTIAL MODELING FOR POSE ESTIMATION Fabio Cuzzolin, Computer Science Department, UCLA; Ruggero Frezza, DEI, Universita’ di PadovaFabio Cuzzolin, Computer Science Department, UCLA; Ruggero Frezza, DEI, Universita’ di Padova

EVIDENTIAL MODELING FOR POSE ESTIMATIONEVIDENTIAL MODELING FOR POSE ESTIMATION Fabio Cuzzolin, Computer Science Department, UCLA; Ruggero Frezza, DEI, Universita’ di PadovaFabio Cuzzolin, Computer Science Department, UCLA; Ruggero Frezza, DEI, Universita’ di Padova

44ndnd INTERNATIONAL SYMPOSIUM ON IMPRECISE PROBABILITIES AND THEIR APPLICATIONS, ISIPTA’05 INTERNATIONAL SYMPOSIUM ON IMPRECISE PROBABILITIES AND THEIR APPLICATIONS, ISIPTA’05 Carnegie Mellon University, Pittsburgh, USA, July 20-23 2005Carnegie Mellon University, Pittsburgh, USA, July 20-23 2005

TRAINING A BOTTOM-UP MODELTRAINING A BOTTOM-UP MODEL

• the collection of approximate parameter space and feature spaces, linked by the refining maps learned in the training stage form an evidential model of the object

• the evidential model can then be used thereafter to estimate the pose of the body when new images become available

BOTTOM LINESBOTTOM LINES

• we want to estimate the we want to estimate the configurationconfiguration (pose) of an (pose) of an unknownunknown object from a sequence of images object from a sequence of images

• model-free pose estimation requires to model-free pose estimation requires to build maps from build maps from image measurements (features) to posesimage measurements (features) to poses

• different features have to be different features have to be integratedintegrated to increase to increase accuracy and robustness of the estimationaccuracy and robustness of the estimation

• the the evidential modelevidential model provides such a framework provides such a framework

MODEL-FREE POSE ESTIMATIONMODEL-FREE POSE ESTIMATION

pose estimation: reconstruction of the actual pose of a moving object by processing the sequence of images taken during its motion

model-based pose estimation: a kinematic model of the body is known and used to help the estimation

• the estimate of component 9 of the pose vector shows a neat improvement when using the comprehensive model

• the silhouette of the body of interest is detected by colorimetric analysis• the bounding box containing the object is found• feature vector = collection of the coordinates of the vertices of the box

• example: Rehg and Kanade• pose = angles between links of the fingers

12

3

• TRAINING : the body moves in front of the camera(s), while a sequence of poses is provided by a motion capture system. Then

• some features are computed from the images• these feature sequences are passed to an HMM with n states yielding:

• the approximate feature spaces• the maps between features and poses

• two views acquired through DV cameras

1

• ESTIMATION : the body performs new movements in front of the camera(s), and for each available image

• the features are computed as before• the likelihoods of the features are transformed into belief functions• those measurement functions are projected onto Q and combined• a pointwise estimate of the object pose is computed by pignistic or relative plausibility transformation

2

• comparison between visual estimate and real view

Tk

kIkpI..1

)()(ˆˆ