My project was about Detection Theory. See my poster
Several cameras scan an unknown map, to find targets. The detection algorithms extract feature points(edges, corners...) and process a binary representation of the possible targets location. False alarm probabilities or prior target distribution are unknown. Three sub problems solved:
- Probabilities problem
- Two independent observers with a unknown probability to say the truth focus on the same target reporting. Unknown prior target distribution. Errors made by the observers (noise) independent of the target presence.
- Goal: How find out the number of targets? How far each observer can be trusted?
- Simultaneous localisation and mapping (SLAM)
- A pltatform moves upon a binary unknown pattern.
Assumptions: - rigid motion of the platform (translations only),
- measurement model of 1. - Goal: How simultaneously recover the pattern and the platform position?
left video is measurements, right one is the posterior probability map to have targets; the rebuilt map.
A new bayesian filter is introduced:
- Targets tracking (Applications) Three cameras in a cluttered environnent, with occlusion Ponctual targets, brownian movment. Targets can appear or disappear Goal: Track targets
To go deeper, see my report
.