Back to Tao Zhao's Research Page

Bayesian Human Segmentation in Crowded Situations

Tao Zhao and Ram Nevatia
{taozhao|nevatia} (at) usc (o) edu
CVPR 2003 (IEEE Conference on Computer Vision and Pattern Recognition), Madison, Wisconsin, Jun., 2003.

Abstract

Problem of segmenting individual humans in crowded situations from stationary video camera sequences is exacerbated by object inter-occlusion. We pose this problem as a ”°model-based segmentation”± problem in which human shape models are used to interpret the foreground in a Bayesian framework. The solution is obtained by using an efficient Markov chain Monte Carlo (MCMC) method which uses domain knowledge as proposal probabilities. Knowledge of various aspects including human shape, human height, camera model, and image cues including human head candidates, foreground/background separation are integrated in one theoretically sound framework. We show promising results and evaluations on some challenging data.

Results on two sequences

Results on Seq1 (Topping):

Selected frames:

The results of all the frames are assembled into a video file (MPEG4 AVI): Topping_IO.avi (12M, high res), Topping_IO_low.avi (5M, low res). However, since the results are generated frame by frame (no temporal consistency is enforced), the readers are requested to view the video frame by frame and not to rely on their motion perceptions.

Results on Seq2 (Commons):

Selected frames:

All frames video file: Commons_IO.avi (8.7M, high res), Commons_IO_low.avi (3.5M, low res).

Performance Evaluation on two sequences

ToppingCommons
valid humans84666726
correct detections78816243
missed-detections585483
false alarms29112
detection rate93.09%92.82%
false alarm rate3.43%0.18%

Histogram of the number of humans per blob as a measurement of the complexity of the data.
Topping
Commons

Download PDF file of the paper.