Online multi-person tracking based on global sparse collaborative representations

Abstract

Multi-person tracking is still a challenging problem due to recurrent occlusion, pose variation and similar appearances between people. Inspired by the success of sparse representations in single object tracking and face recognition, we propose in this paper an online tracking by detection framework based on collaborative sparse representations. We argue that collaborative representations can better differentiate people compared to target-specific models and therefore help to produce a more robust tracking system. We also show that despite the size of the dictionaries involved, these representations can be efficiently computed with large-scale optimization techniques to get a near real-time algorithm. Experiments show that the proposed approach compares well to other recent online tracking systems on various datasets.

Collaboration and spatialization for an efficient multi-person tracking via sparse representations

“…The element to be classified is then represented in a collaborative way among all classes, and its class is estimated as the one that minimizes a residual reconstruction error [14]. In this paper, we propose to compute the affinity values for all the track-detection pairs using this kind of classifier with a local description of the targets, building on the work done in [4] which considers only holistic descriptions.…”

Improving Multi-frame Data Association with Sparse Representations for Robust Near-online Multi-object Tracking

“…However, as many specific and independent models as the number of targets are necessary. In contrast, in [3,27], a single dictionary is shared by all targets and collaborative representations are used to better discriminate them. All these MOT approaches are using a twoframe data association in an online fashion and thus cannot reconsider wrong associations when further information comes and contradicts them.…”

“…Our approach is inspired by [3,27,28], but instead of relying on sparse representations induced by the standard l 1 norm, we design a sparsity-inducing norm, based on a weighted l ∞,1 norm, more suited for a multi-frame data association problem.…”

“…Each candidate for the new target location is then represented by a sparse linear combination of the dictionary elements, the best reconstruction error being used as the selection criterion. However, only a few recent methods have considered extending these models for online MOT systems [3,27,28].…”

“…In the literature, the problem is addressed by a large variety of approaches, from online (or single-scan) techniques [1][2][3][4] where only the previous frames are considered, to offline approaches using past and future frames. Among offline techniques, global approaches perform the data association over all the frames simultaneously or by batch [5][6][7][8][9][10][11][12][13][14][15], whereas sliding window (a.k.a.…”