SpaRTA Tracking Across Occlusions via Partitioning of 3D Clouds of Points.

Andrea Cavagna,Leonardo Parisi,Federico Ricci-Tersenghi,Stefania Melillo

doi:10.1109/tpami.2019.2946796

Abstract

Any 3D tracking algorithm has to deal with occlusions: multiple targets get so close to each other that the loss of their identities becomes likely; hence, potentially affecting the very quality of the data with interrupted trajectories and identity switches. Here, we present a novel tracking method that addresses the problem of occlusions within large groups of featureless objects by means of three steps: i) it represents each target as a cloud of points in 3D; ii) once a 3D cluster corresponding to an occlusion occurs, it defines a partitioning problem by introducing a cost function that uses both attractive and repulsive spatio-temporal proximity links; and iii) it minimizes the cost function through a semi-definite optimization technique specifically designed to cope with the presence of multi-minima landscapes. The algorithm is designed to work on 3D data regardless of the experimental method used: multicamera systems, lidars, radars, and RGB-D systems. By performing tests on public data-sets, we show that the new algorithm produces a significant improvement over the state-of-the-art tracking methods, both by reducing the number of identity switches and by increasing the accuracy of the estimated positions of the targets in real space.

Highlights

TRACKING large groups of targets in 3D space is a challenging topic, which is relevant in the field of turbulence [1], collective animal behavior [2], [3] and social sciences [4], [5], [6] as well as in robotics [7] and autonomous mobility [8]
We report the MOTA (Multiple Object Tracking Accuracy) and the number of switches of identities (IDS), the percentage of mostly tracked (MT) and mostly lost (ML) trajectories corresponding to groundtruth trajectories which are correctly reconstructed respectively for more than the 80 percent and for less than the 20 percent of their time length, the number of tracks fragments (FM) corresponding to the number of times that a groundtruth trajectory, correctly reconstructed, is interrupted
In the sparse dataset the percentage of MT obtained by SpaRTA (96.1 percent) is comparable with the 3 Single Point (SIP)-RT methods MHT, SDD–MHT and CP(LDQD), while on the dense dataset SpaRTA produces the highest percentage of MT (91.1 percent)

Summary

Introduction

TRACKING large groups of targets in 3D space is a challenging topic, which is relevant in the field of turbulence [1], collective animal behavior [2], [3] and social sciences [4], [5], [6] as well as in robotics [7] and autonomous mobility [8]. The crucial point of all tracking algorithms is how to handle occlusions that arise every time that two or more objects get too close in 3D space to be detected as multiple targets. This kind of ambiguities are severe when dealing with featureless objects (objects that cannot be identified by any feature such as shape or color) and with large and dense groups of targets, where the chance to get in 3D proximity is high. Occlusions hinder in a twofold way the quality of the retrieved trajectories: loss of one or more of the targets involved into the occlusions and a potential switch of identities

Methods

Results

Conclusion