Capturing the Sporting Heroes of Our Past by Extracting 3D Movements from Legacy Video Content

Abstract

Sports are a key part of cultural identity, and it is necessary to preserve them as important intangible Cultural Heritage, especially the human motion techniques specific to individual sports. In this paper we present a method for extracting 3D athlete motion from video broadcast sources, providing an important tool for preserving the heritage represented by these movements. Broadcast videos include camera motion, multiple player interaction, occlusions and noise, presenting significant challenges to solve the reconstruction. The approach requires initial definition of some key-frames and setting of 2D key-points in those frames manually. Thereafter an automatic process estimates the poses and positions of the players in the key-frames, and in the frames between key-frames, taking into account collisions with the environment and human kinematic constraints. Initial results are extremely promising and this data could be used to analyze the sport's evolution over time, or even to generate animations for interactive applications.

Virtual Reality for the Preservation and Promotion of Historical Real Tennis

“…Skublewska-Paszkowska et al [21] conducted a systematic review of the literature to identify relevant studies related to 3D technologies and intangible cultural heritage preservation. They specifically review two studies related to sport, [10,12]. Both focus on the extraction and reproduction of 3D sports movements, from 2D images.…”

An Overview on Image-Based and Scanner-Based 3D Modeling Technologies

“…In addition to the conventional applications of 3D modeling in the field of culture (3D modeling and geometric documentation of cultural heritage sites/buildings/monuments and 3D modeling of objects/relics of high cultural value), in recent years, emphasis has also been placed on 3D modeling of intangible cultural heritage [ 102 ]. Such examples concern the 3D modeling of dances through the 3D recording of dancers’ dance movements [ 103 , 104 ], the 3D recording of athletes’ movements, in some cases in combination with the surrounding environment and/or objects [ 105 ], and the 3D motion recording for traditional crafts [ 106 , 107 ]. In addition, 3D modeling applications of underwater spaces or cultural heritage objects are of special interest [ 108 ].…”

Contextualised learning‐free three‐dimensional body pose estimation from two‐dimensional body features in monocular images

“…Learning‐free approaches model the human body as a set of kinematic chains and reconstruct the 3D pose given the 2D joint projections through kinematic constraints [10–14]. Taylor [10] recovers the entire set of 3D pose solutions by considering known skeletal sizes and human input to resolve the depth ambiguities at each joint, which are not available in our context.…”

“…The key‐frame in‐betweens are estimated by a combination of vision‐based tracking and interpolations. Goenetxea et al [14] extended this approach by including the kinematic correlations between all the key‐frames during the 3D reconstruction, obtaining more coherent estimations throughout the sequence.…”

“…Monocular 3D human pose estimation methods can be broadly categorised into learning-free [10][11][12][13][14] and learning-based [15][16][17]. Learning-free techniques rely on kinematic constraints on pose and movement to solve the inherent ambiguity in monocular human pose reconstruction, while learning-based methods rely on a training stage that associates 3D human poses to their corresponding 2D projections.…”