In order to track all persons in a scene, the tracking-by-detection paradigm
has proven to be a very effective approach. Yet, relying solely on a single
detector is also a major limitation, as useful image information might be
ignored. Consequently, this work demonstrates how to fuse two detectors into a
tracking system. To obtain the trajectories, we propose to formulate tracking
as a weighted graph labeling problem, resulting in a binary quadratic program.
As such problems are NP-hard, the solution can only be approximated. Based on
the Frank-Wolfe algorithm, we present a new solver that is crucial to handle
such difficult problems. Evaluation on pedestrian tracking is provided for
multiple scenarios, showing superior results over single detector tracking and
standard QP-solvers. Finally, our tracker ranks 2nd on the MOT16 benchmark and
1st on the new MOT17 benchmark, outperforming over 90 trackers.

本文介绍了如何将两种探测器融合到一个跟踪系统中，以解决单个探测器的局限性，并且提出了一种基于加权图标记问题的追踪算法，该算法通过 Frank-Wolfe 算法实现，经实验证明在行人跟踪的多种情景下，效果优于单个探测器跟踪和标准 QP-solvers，相比超过 90％的跟踪器，在 MOT16 基准测试中排名第二，在新的 MOT17 基准测试中排名第一。

头部和全身探测器的融合用于多物体追踪

Fusion of Head and Full-Body Detectors for Multi-Object Tracking

We consider the two block stochastic block model on $n$ nodes with
asymptotically equal cluster sizes. The connection probabilities within and
between cluster are denoted by $p_n:=\frac{a_n}{n}$ and $q_n:=\frac{b_n}{n}$
respectively. Mossel et al.(2012) considered the case when $a_n=a$ and $b_n=b$
are fixed. They proved the probability models of the stochastic block model and
that of Erd{\"o}s-R{\'e}nyi graph with same average degree are mutually
contiguous whenever $(a-b)^2<2(a+b)$ and are asymptotically singular whenever
$(a-b)^2>2(a+b)$. Mossel et al.(2012) also proved that when $(a-b)^2<2(a+b)$ no
algorithm is able to find an estimate of the labeling of the nodes which is
positively correlated with the true labeling. It is natural to ask what happens
when $a_n$ and $b_n$ both grow to infinity. We prove that their results extend
to the case when $a_n=o(n)$ and $b_n=o(n)$. We also consider the case when
$\frac{a_n}{n} \to p \in (0,1)$ and $(a_n-b_n)^2= \Theta(a_n+b_n)$. Observe
that in this case $\frac{b_n}{n} \to p$ also. We show that here the models are
mutually contiguous if $(a_n-b_n)^2< 2(1-p)(a_n+b_n)$ and they are
asymptotically singular if $(a_n-b_n)^2 > 2(1-p)(a_n+b_n)$. Further we also
prove it is impossible find an estimate of the labeling of the nodes which is
positively correlated with the true labeling whenever $(a_n-b_n)^2<
2(1-p)(a_n+b_n)$. The results of this paper justify the negative part of a
conjecture made in Decelle et al.(2011) for dense graphs.

本研究通过渐近分析证明，若随机块模型中连接概率符合一定条件，则节点标签的真实估计无法与真实标签正相关，这一结论适用于固定连接概率以及随着节点数增加而连接概率下降的情况。