Tracking Multiple Persons Based on a Variational Bayesian Model

MOT16-01 MOT16-03 MOT16-06 MOT16-07 MOT16-08 MOT16-12 MOT16-14

Short name:

OVBT

Benchmark:

Description:

An online variational
Bayesian model for multiple-person tracking is proposed. This yields a
variational expectation-maximization (VEM) algorithm. The computa-
tional efficiency of the proposed method is due to closed-form expressions
for both the posterior distributions of the latent variables and for the es-
timation of the model parameters. A stochastic process that handles
person birth and person death enables the tracker to handle a varying
number of persons over long periods of time.

Hardware:

2.53GHz

Detector:

Public

Processing:

Online

Last submitted:

August 16, 2016 (1 year ago)

Published:

July 16, 2016 at 12:07:44 CET

Submissions:

3

Open source:

No

Project page / code:

n/a

Reference:

Y. Ban, S. Ba, X. Alameda-Pineda, R. Horaud. Tracking Multiple Persons Based on a Variational Bayesian Model. In BMTT 2016, .

Benchmark performance:

MOTAMOTPFAFMTMLFPFNID Sw.FragSpecificationsDetector
38.475.41.97.5 % 47.3 % 11,51799,4631,3212,1402.53GHzPublic

Detailed performance:

Sequence MOTA MOTP FAF GT MT ML FP FN ID Sw Frag
MOT16-0123.971.41.52313.0 % 39.1 % 6964,1373589
MOT16-0346.975.74.114817.6 % 20.3 % 6,17348,6316891,184
MOT16-0632.773.20.52213.6 % 58.4 % 5627,073124183
MOT16-0733.673.32.2549.3 % 35.2 % 1,0779,605158272
MOT16-0824.678.41.7633.2 % 41.3 % 1,06611,402150177
MOT16-1232.876.70.98610.5 % 52.3 % 7664,7496380
MOT16-1418.174.51.61642.4 % 61.6 % 1,17713,866102155

Raw data:


OVBT
MC