Articolo in rivista, 2015, ENG, 10.1109/TPAMI.2015.2414427

GReTA-A Novel Global and Recursive Tracking Algorithm in Three Dimensions

Alessandro Attanasi; Andrea Cavagna; Lorenzo Del Castello; Irene Giardina; Asja Jelic; Stefania Melillo; Leonardo Parisi; Fabio Pellacini; Edward Shen; Edmondo Silvestri; Massimiliano Viale

SISTeMA ITS s.r.l.00153, Rome, Italy; Dipartimento di Fisica, Università Sapienza, 00185, Rome, Italy and also with the Istituto Sistemi Complessi, Consiglio Nazionale delle Ricerche, u.o.s. Sapienza, 00185 Rome, Italy; Dipartimento di Informatica, Università Sapienza,00198 Rome, Italy and with the Istituto Sistemi Complessi, Consiglio Nazionale delle Ricerche, u.o.s. Sapienza, 00185 Rome, Italy; Dipartimento di Informatica, Università Sapienza 00198 Rome, Italy; Abdus Salam International Centre for Theoretical Physics34014, Trieste, Italy; Bublcam Technology Inc., Toronto, Canada

Tracking multiple moving targets allows quantitative measure of the dynamic behavior in systems as diverse as animal groups in biology, turbulence in fluid dynamics and crowd and traffic control. In three dimensions, tracking several targets becomes increasingly hard since optical occlusions are very likely, i.e., two featureless targets frequently overlap for several frames. Occlusions are particularly frequent in biological groups such as bird flocks, fish schools, and insect swarms, a fact that has severely limited collective animal behavior field studies in the past. This paper presents a 3D tracking method that is robust in the case of severe occlusions. To ensure robustness, we adopt a global optimization approach that works on all objects and frames at once. To achieve practicality and scalability, we employ a divide and conquer formulation, thanks to which the computational complexity of the problem is reduced by orders of magnitude. We tested our algorithm with synthetic data, with experimental data of bird flocks and insect swarms and with public benchmark datasets, and show that our system yields high quality trajectories for hundreds of moving targets with severe overlap. The results obtained on very heterogeneous data show the potential applicability of our method to the most diverse experimental situations.

IEEE transactions on pattern analysis and machine intelligence 37 (12), pp. 2451–2463

Keywords

3D, tracking, branching, divide and conquer, global optimization, multi-object, multi-path, recursion, tracking

CNR authors

Melillo Stefania, Del Castello Lorenzo, Parisi Leonardo, Giardina Irene Rosana, Viale Massimiliano, Cavagna Andrea

CNR institutes

ISC – Istituto dei sistemi complessi