GMCP-Tracker: Global Multi-object Tracking Using Generalized Minimum Clique Graphs

Abstract

Abstract. Data association is an essential component of any human tracking system. The majority of current methods, such as bipartite matching, incorporate a limited-temporal-locality of the sequence into the data association problem, which makes them inherently prone to IDswitches and difficulties caused by long-term occlusion, cluttered background, and crowded scenes. We propose an approach to data association which incorporates both motion and appearance in a global manner. Unlike limited-temporal-locality methods which incorporate a few frames into the data association problem, we incorporate the whole temporal span and solve the data association problem for one object at a time, while implicitly incorporating the rest of the objects. In order to achieve this, we utilize Generalized Minimum Clique Graphs to solve the optimization problem of our data association method. Our proposed method yields a better formulated approach to data association which is supported by our superior results. Experiments show the proposed method makes significant improvements in tracking in the diverse sequences of

Tracking Multiple People Online and in Real Time

“…The work most similar to ours is that of Zamir et al [13], who formulate multiperson tracking as a sequential Generalized Minimum Clique Problem (GMCP) decomposition of a complete graph of observations. As it is common in the literature, they also decompose computation into two stages for computational efficiency, and can incorporate hard constraints on co-identity.…”

“…Methods that use stronger and more comprehensive evidence have demonstrated superior performance. These methods consider evidence from all observation pairs [13], observation triplets [14] or higher order relationships [15][16][17][18]. The better performance however comes at a cost of increased computational complexity due to the problem's combinatorial nature.…”

“…Despite these difficulties, several strategies have been proposed in the literature to address computational complexity, such as (i) approximating the problem formulation, (ii) approximating the solution of the problem [1,3,13,19,20], (iii) limiting the number of edges in the observation graph [9,10], (iv) relaxing constraints in the BIP [21][22][23] or (v) solving the problem in stages for efficiency [3,11,13,19,20].…”

“…We evaluate our algorithm on three standard single-camera datasets for multiperson tracking: PETS2009 [30], Town Center [20] and Parking Lot [13]. We used the PETS2009-S2L1 View 1 sequence, which has a resolution of 768 x 576 pixels and consists of 798 frames at 7 fps (117 seconds).…”

“…MOTA IDsw Berclaz [10] 80.00 28 Shitrit [9] 81.46 19 Andriyenko [19] 81.84 15 Henriques [6] 87.95 10 Izadinia [11] 90.70 -Zamir [13] 91.50 8 Ours 93.34 1…”