Philip Kelly

Evaluating a dancer's performance using kinect-based skeleton tracking

In this work, we describe a novel system that automatically evaluates dance performances against a gold-standard performance and provides visual feedback to the performer in a 3D virtual environment. The system acquires the motion of a performer via Kinect-based human skeleton tracking, making the approach viable for a large range of users, including home enthusiasts. Unlike traditional gaming scenarios, when the motion of a user must by kept in synch with a pre-recorded avatar that is displayed on screen, the technique described in this paper targets online interactive scenarios where dance choreographies can be set, altered, practiced and refined by users. In this work, we have addressed some areas of this application scenario. In particular, a set of appropriate signal processing and soft computing methodologies is proposed for temporally aligning dance movements from two different users and quantitatively evaluating one performance against another.

Robust pedestrian detection and tracking in crowded scenes

In this paper, a robust computer vision approach to detecting and tracking pedestrians in unconstrained crowded scenes is presented. Pedestrian detection is performed via a 3D clustering process within a region-growing framework. The clustering process avoids using hard thresholds by using bio-metrically inspired constraints and a number of plan-view statistics. Pedestrian tracking is achieved by formulating the track matching process as a weighted bipartite graph and using a Weighted Maximum Cardinality Matching scheme. The approach is evaluated using both indoor and outdoor sequences, captured using a variety of different camera placements and orientations, that feature significant challenges in terms of the number of pedestrians present, their interactions and scene lighting conditions. The evaluation is performed against a manually generated groundtruth for all sequences. Results point to the extremely accurate performance of the proposed approach in all cases.

TennisSense: A platform for extracting semantic information from multi-camera tennis data

In this paper, we introduce TennisSense, a technology platform for the digital capture, analysis and retrieval of tennis training and matches. Our algorithms for extracting useful metadata from the overhead court camera are described and evaluated. We track the tennis ball using motion images for ball candidate detection and then link ball candidates into locally linear tracks. From these tracks we can infer when serves and rallies take place. Using background subtraction and hysteresis-type blob tracking, we track the tennis players positions. The performance of both modules is evaluated using ground-truthed data. The extracted metadata provides valuable information for indexing and efficient browsing of hours of multi-camera tennis footage and we briefly illustrative how this data is used by our tennis-coach playback interface.

An advanced virtual dance performance evaluator

The ever increasing availability of high speed Internet access has led to a leap in technologies that support real-time realistic interaction between humans in online virtual environments. In the context of this work, we wish to realise the vision of an online dance studio where a dance class is to be provided by an expert dance teacher and to be delivered to online students via the web. In this paper we study some of the technical issues that need to be addressed in this challenging scenario. In particular, we describe an automatic dance analysis tool that would be used to evaluate a students performance and provide him/her with meaningful feedback to aid improvement.

Analysis of the 5 iron golf swing when hitting for maximum distance

Abstract Most previous research on golf swing mechanics has focused on the driver club. The aim of this study was to identify the kinematic factors that contribute to greater hitting distance when using the 5 iron club. Three-dimensional marker coordinate data were collected (250 Hz) to calculate joint kinematics at eight key swing events, while a swing analyser measured club swing and ball launch characteristics. Thirty male participants were assigned to one of two groups, based on their ball launch speed (high: 52.9 ± 2.1 m · s−1; low: 39.9 ± 5.2 m · s−1). Statistical analyses were used to identify variables that differed significantly between the two groups. Results showed significant differences were evident between the two groups for club face impact point and a number of joint angles and angular velocities, with greater shoulder flexion and less left shoulder internal rotation in the backswing, greater extension angular velocity in both shoulders at early downswing, greater left shoulder adduction angular velocity at ball contact, greater hip joint movement and X Factor angle during the downswing, and greater left elbow extension early in the downswing appearing to contribute to greater hitting distance with the 5 iron club.

A Sensing Platform for Physiological and Contextual Feedback to Tennis Athletes

In this paper we describe our work on creating a multimodal sensing platform for providing feedback to tennis coaches and players. The platform includes a fixed installation around a tennis court consisting of a video camera network and a localisation system as well as wearable sensing technology deployed to individual athletes. We describe the various components of this platform and explain how we can capture synchronised multi-modal sensor data streams for games or training sessions. We then describe the content-based retrieval system we are building to facilitate the development of novel coaching tools. We provide some examples of the queries that the system can support, where these queries are chosen to be suitably expressive so as to reflect a coach’s complex information needs regarding tennis-related performance factors.

A Framework for Evaluating Stereo-Based Pedestrian Detection Techniques

Automated pedestrian detection, counting, and tracking have received significant attention in the computer vision community of late. As such, a variety of techniques have been investigated using both traditional 2-D computer vision techniques and, more recently, 3-D stereo information. However, to date, a quantitative assessment of the performance of stereo-based pedestrian detection has been problematic, mainly due to the lack of standard stereo-based test data and an agreed methodology for carrying out the evaluation. This has forced researchers into making subjective comparisons between competing approaches. In this paper, we propose a framework for the quantitative evaluation of a short-baseline stereo-based pedestrian detection system. We provide freely available synthetic and real-world test data and recommend a set of evaluation metrics. This allows researchers to benchmark systems, not only with respect to other stereo-based approaches, but also with more traditional 2-D approaches. In order to illustrate its usefulness, we demonstrate the application of this framework to evaluate our own recently proposed technique for pedestrian detection and tracking.

A virtual coaching environment for improving golf swing technique

As a proficient golf swing is a key element of success in golf, many golfers make significant effort improving their stroke mechanics. In order to help enhance golfing performance, it is important to identify the performance determining factors within the full golf swing. In addition, explicit instructions on specific features in stroke technique requiring alterations must be imparted to the player in an unambiguous and intuitive manner. However, these two objectives are difficult to achieve due to the subjective nature of traditional coaching techniques and the predominantly implicit knowledge players have of their movements. In this work, we have developed a set of visualisation and analysis tools for use in a virtual golf coaching environment. In this virtual coaching studio, the analysis tools allow for specific areas require improvement in a players 3D stroke dynamics to be isolated. An interactive 3D virtual coaching environment then allows detailed and unambiguous coaching information to be visually imparted back to the player via the use of two virtual human avatars; the first mimics the movements performed by the player; the second takes the role of a virtual coach, performing ideal stroke movement dynamics. The potential of the coaching tool is highlighted in its use by sports science researchers in the evaluation of competing approaches for calculating the X-Factor, a significant performance determining factor for hitting distance in a golf swing.

Game, shot and match: Event-based indexing of tennis

Identifying events in sports video offers great potential for advancing visual sports coaching applications. In this paper, we present our results for detecting key events in a tennis match. Our overall goal is to automatically index a complete tennis match into all the main tennis events, so that a match can be recorded using affordable visual sensing equipment and then be automatically indexed into key events for retrieval and editing. The tennis events detected in this paper are a tennis game, a change of end and a tennis serve — all of which share temporal commonalities. There are of course other events in tennis which we aim to index in our overall indexing system, but this paper focuses solely on the aforementioned tennis events. This paper proposes a novel approach to detect key events in an instrumented tennis environment by analysing a players location and the visual features of a player.

Combining inertial and visual sensing for human action recognition in tennis

In this paper, we present a framework for both the automatic extraction of the temporal location of tennis strokes within a match and the subsequent classification of these as being either a serve, forehand or backhand. We employ the use of low-cost visual sensing and low-cost inertial sensing to achieve these aims, whereby a single modality can be used or a fusion of both classification strategies can be adopted if both modalities are available within a given capture scenario. This flexibility allows the framework to be applicable to a variety of user scenarios and hardware infrastructures. Our proposed approach is quantitatively evaluated using data captured from elite tennis players. Results point to the extremely accurate performance of the proposed approach irrespective of input modality configuration