3D Tracking Using Occupancy Grids
Post-doctoral fellow
Edmond Boyer Jean-Sebastien Franco
This 1 year project deals with multiple camera environments, and considers the problem of tracking moving objects in such environments. We have recently proposed a method that computes occupancy probabilities at discrete locations in a 3D space. These probabilities are obtained by fusing information from several 2D images, typically silhouette information, acquired at each instant. The result is a 3D grid of voxels having real values which encode the probability that a voxel belongs to a moving object (see reference [1] below for more details).
The goal of this project is to develop a 3D tracking method based on these grids. To this purpose, we will consider grid sequences over time and how to impose temporal consistency over these sequences. An intuitive first solution is to consider objects as single points inside the grid (e.g. probability maxima for instance) and then to track the trajectories of these points using standard filtering methods. A more challenging issue is to consider the occupancy grid itself and how to account for temporal consistency within the grid, through a dynamic motion model. The interest is to propose robust solutions for tracking moving objects such as persons at different levels of details, from positions to complete shapes.
This research will be conducted at the INRIA Rhone-Alpes (Grenoble-France) within the Perception team, and developments will be made in the context of the GrImage platform: a multi-camera acquisition platform, with a dedicated PC cluster, and a multi-projector display wall.
References:
[1] Fusion of Multi-View Silhouette Cues Using a Space Occupancy Grid J.-S. Franco and E. Boyer International Conference on Computer Vision (ICCV’05)
[2] Fast, Integrated Person Tracking and Activity Recognition with Plan-View Templates from a Single Stereo Camera M. Harville and D. Li IEEE Conference on Vision and Pattern Recognition (CVPR’04)
Start date: 1 February 2006
Contact person: Edmond BOYER
Deadline: 15 January 2006

