Spatio-temporal target-measure association using an adaptive geometrical approach

Data association is of crucial importance to improve target tracking performance in many complex visual environments (non-linear dynamics, occlusions, etc). Usually, association effectiveness is based on prior information and observation category. However, association becomes difficult if targets are similar. Problems also arise in cases of missing data, complex motions or deformations over time. To remedy, we propose a new method for data association, that uses the evolution of the dynamic model of targets. The main idea is to measure an adaptive geometric accuracy between possible trajectories of targets, by only using positions as information, that constitutes its main advantage.