The main goal is the detection and tracking of agents while they are still some distance away from a particular location, for example a bus station, a pedestrian crossing, a passenger in an airport, or a guest in a lobby.
Detection of human motion and abandoned objects
Popular approaches based on background subtraction use colour information to model each pixel during a training period. Nevertheless, a deep analysis on colour segmentation problems demonstrates that colour is not enough to detect all foreground objects in the image.
Our segmentation procedure is based not only on colour, but also on intensity when the use of colour is not feasible. We also exploit the knowledge obtained from those detected objects which should be incorporated into the background model since they cease their movement. These objects are identified, becoming part of the new background model. Such motionless agents are processed for appearance analysis and agent classification.
Tracking of multiple-people in unconstrained environments
Our tracking system deals with multiple targets whose dynamics are unknown and highly non-linear. They are tracked through outdoor or indoor scenes with complex clutter, which may mimic the target appearances. Moreover, their trajectories can intersect causing complete occlusions.
In addition, the proposed system copes with heavy appearance and shape changes caused by the articulate nature of the humans and the variable background conditions, such as illumination or weather changes.
As a result, target trajectories are obtained, as well as quantitative information about the target state such as their speed or size, and qualitative one, such as whether they are being occluded, grouping or are entering or exiting the scene.