Citation Context

...oor fields from the visual data: a ‘Static Floor Field’ (SFF), a ‘Boundary Floor Field’ (BFF), and a ‘Dynamic Floor Field’ (DFF). 2 Related Work Tracking is a popular research area in computer vision =-=[16, 6]-=-. In this section, we briefly review tracking algorithms that were designed specifically for crowdedFloor Fields for Tracking in High Density Crowd Scenes 3 scenes. Interested readers are referred to...

Citation Context

...ing has been studied intensively. For example, Kalman filtering has been a classic scheme for object tracking. Recently, particle filtering has been popular for tracking multiple objects such as ants =-=[2]-=- with complex interactions. Particle filtering has also been studied for tracking hockey players [3] in which object interaction is not explicitly modeled. Bayesian networks have been applied to optim...

Citation Context

...ntities and interactions between objects can not be easily modeled in formal (and algorithmic) terms. For these reasons, many adopt a rigorous formulation of the MOT problem using a joint state space =-=[9, 14, 5, 6, 16, 15]-=-, where object interactions can be properly defined. However, sampling from a joint state space can quickly become inefficient as the dimension of the space increases. Recently, finding an efficient M...

Citation Context

...mework. Brostow and Cipolla [5] present a probabilistic framework for the clustering of feature point trajectories to detect individual pedestrians in a crowd. These and other similar approaches like =-=[28]-=-, [36], [47], [54] skip the modeling of articulations in favor of appearance models trained for specific unoccluded views of their respective subjects. As a result, they are challenged by fully and pa...

Citation Context

...ame target, this will be done automatically in a probabilistic framework. This in essence is the approach of this paper. Over the last couple of years, many algorithms and results have been presented =-=[7, 3]-=- with regard to the problem of multiple object tracking. Prevalent are algorithms based on Kalman filtering [15, 5], advanced techniques of particle filtering [14, 10, 9, 2] and multiple-hypothesis tr...

Citation Context

...6]. In this paper, we address the problem as one of approximate inference in a dynamical graphical model, using particle filtering (PF) techniques [11], [18], building on recent advances in the field =-=[22]-=-. For a state-space model, a PF recursively approximates the filtering distribution of states given observations, using a dynamical model, an observation model, and sampling techniques, by predicting ...

Citation Context

... The readers interested in a detailed review of the state of the art in tracking are referred to a recent survey by Yilmaz et al. [7]. An interesting body of work tries to track sparse crowds of ants =-=[4]-=- and people [8], hockey players [6], crowds of densely packed people [5, 11, 12], a dense flock of bats [2], and biological cells [3]. In their work, Brostow et al. [8] tracked and clustered feature p...

Citation Context

... are shown as dashed lines in the same event. photo: ψ u � 1 if xi �= x i, j(xi,xj) = j; τ if xi = x j. This type of MRF constraints has been used before in different applications, including tracking =-=[7]-=- and the face clustering work of Song and Leung [22]. Setting τ = 0 enforces the constraint strictly, however in our experiments τ = 0.25 performed slightly better. While the the model above allows us...

Citation Context

...en objects. As a result, much work has been focused on adopting a rigorous Bayesian joint state-space formulation to the problem, where object interactions can be explicitly defined [20], [39], [15], =-=[17]-=-, [44], [32]. However, sampling from a joint state-space can quickly become inefficient, as the dimension of the space increases when more people are added [20]. Recent work has concentrated on using ...