Vision-based group-behavior evaluation in delivery simulation training: a view-independent approach

This paper aims at generating an automated way to evaluate the group-behavior of trainees in a delivery simulation course using video-processing techniques. The paper is composed of three interacting, but clearly separated stages: moving people segmentation, multiple people tracking and group-behavior analysis. At the segmentation stage, the combination of the Gaussian Mixture Model (GMM) and the Dynamic Markov Random Fields (DMRF) technique helps to precisely extract the foreground pixels. At the tracking stage, we concentrate on solving human-occlusion problem based on silhouette data and a non-linear regression model. Our model e®ectively transfers the person location prob- lem during the occlusion into the ¯nding of the local maximum points on a smooth curve, so that visual persons in the partial or complete occlusion can still be localized. At group-behavior analysis stage, we apply a Hidden Markov Model (HMM) to infer human behavior, in which useful visual features are upgraded to real-world domain by employing a homography mapping-based camera calibration algorithm. Exploiting the camera information results in a robust recognition of the behavior, since the features extracted in the real-world domain are invariant to the changes of camera views. The comparison results reveal that the tracking correctness of our method is much higher than the mean-shift algorithm and slightly lower than a particle-¯lter, however, with a bene¯t of being a factor of 10-15 faster in computing.