Sport players detection and tracking with a mixed network of planar and omnidirectional cameras

Abstract

A generic approach is presented to detect and track people with a network of fixed and omnidirectional cameras given severely degraded foreground silhouettes. The problem is formulated as a sparsity constrained inverse problem. A dictionary made of atoms representing the silhouettes of a person at a given location is used within the problem formulation. A reweighted scheme is considered to better approximate the sparsity prior.Although the framework is generic to any scene, the focus of this paper is to evaluate the performance of the proposed approach on a basketball game. The main challenges come from the players' behavior, their similar appearance, and the mutual occlusions present in the views. In addition, the extracted foreground silhouettes are severely degraded due to the polished floor reflecting the players, and the strong shadow present in the scene. We present qualitative and quantitative results with the APIDIS dataset as part of the ICDSC sport challenge. 1

Sparsity Driven People Localization with a Heterogeneous Network of Cameras

“…Without the re-weighted scheme, the solution is not sparse enough. It leads to very high number of false positives as presented in [1,2].…”

“…In order to linearize (2) and (3), we first remove the quantization operator Q from their fidelity terms. This amounts to consider possible object occlusions has an additional noise on the measured silhouettes, increasing therefore the value of the desired residual error ε or the maximum number of people ε p in (2) and (3) respectively.…”

Dempster-Shafer based multi-view occupancy maps

“…Bottomup methods transfer the information in the different camera images to a common reference plane using camera image-floor homographies [1]. Top-down approaches extract occupied ground positions by comparing a generative model of the objects in the scene with the actual foreground silhouettes observed in the camera views [2,3]. Until now, for both approaches the mathematical laws for the fusion of data from different cameras have not been considered explicitly.…”

Optimized information processing in resource-constrained vision systems. From low-complexity coding to smart sensor networks

“…Bottom-up methods transfer the foreground silhouettes from the different camera images to a common reference plane using camera image-floor homographies [Delannay et al, 2009]. Top-down approaches extract occupied ground positions by comparing a generative model of the objects in the scene with the actual foreground silhouettes observed in the camera views [Fleuret et al, 2008, Alahi et al, 2009. Until now, for both approaches the mathematical laws for the fusion of data from different cameras have not been considered explicitly.…”

“…Such maps are important in many applications such as surveillance, smart rooms, video conferencing and sport games analysis. The 2D occupancy sensing systems that are proposed in research are based on a single-camera setup, a multi-camera setup [Delannay et al, 2009, Fleuret et al, 2008, Alahi et al, 2009, pressure sensitive carpets [Clos et al, 2004, Federspiel andMichael, 2005], passive infrared (PIR) sensors [Elwell, 2009, Zhevelev et al, 2009, or active radar/ultrasound/radio beacons [Bahl andPadmanabhan, 2000, McCarthy andMuller, 2005] (or a combination of the previous, e.g. infrared and ultrasonic [Fowler, 2000, Myron et al, 1997, Elwell, 2009).…”