Nikos Katsarakis

Enhanced Human Body Fall Detection Utilizing Advanced Classification of Video and Motion Perceptual Components

The monitoring of human physiological data, in both normal and abnormal situations of activity, is interesting for the purpose of emergency event detection, especially in the case of elderly people living on their own. Several techniques have been proposed for identifying such distress situations using either motion, audio or video data from the monitored subject and the surrounding environment. This paper aims to present an integrated patient fall detection platform that may be used for patient activity recognition and emergency treatment. Both visual data captured from the users environment and motion data collected from the subjects body are utilized. Visual information is acquired using overhead cameras, while motion data is collected from on-body sensors. Appropriate tracking techniques are applied to the aforementioned visual perceptual component enabling the trajectory tracking of the subjects. Acceleration data from the sensors can indicate a fall incident. Trajectory information and subjects visual location can verify fall and indicate an emergency event. Support Vector Machines (SVM) classification methodology has been evaluated using the latter acceleration and visual trajectory data. The performance of the classifier has been assessed in terms of accuracy and efficiency and results are presented.

The AIT 3D Audio / Visual Person Tracker for CLEAR 2007

This paper presents the Athens Information Technology system for 3D person tracking and the obtained results in the CLEAR 2007 evaluations. The system utilizes audiovisual information from multiple acoustic and video sensors. The proposed system comprises a video and an audio subsystem whose results are suitably combined to track the last active speaker. The video subsystem combines in 3D a number of 2D face localization systems, aiming at tracking all people present in a room. The audio subsystem uses an information theoretic metric upon an ensemble of microphones to estimate the active speaker.

3D audiovisual person tracking using Kalman filtering and information theory

This paper proposes a system for tracking people in three dimensions, utilizing audiovisual information from multiple acoustic and video sensors. The proposed system comprises a video and an audio subsystem combined using a Kalman filter. The video subsystem combines in 3D a number of 2D trackers based on a variation of Stauffers adaptive background algorithm with spacio-temporal adaptation of the learning parameters and a Kalman tracker in a feedback configuration. The audio subsystem uses an information theoretic metric upon a pair of microphones to estimate the direction from which sound is arriving from. Combining measurements from a series of pairs the actual coordinate of the speaker in space is derived.

Event detection in athletics for personalized sports content delivery

Broadcasting of athletics is nowadays biased towards running (sprint and longer distances) sports. Personalized content delivery can change that for users that wish to focus on different content. Using a combination of video signal processing algorithms and live information that accompanies the video of large-scale sports like the Olympics, a system can attend to the preferences of users by selecting the most suitable camera view for them.There are two types of camera selection for personalized content delivery. According to the between sport camera selection, the view is changed between two sports, upon the onset of a sport higher up the user preferences than the one currently being delivered. According to the within sport camera selection, the camera is changed to offer a better view of the evolution of the sport, based on the phase it is in. This paper details the video processing algorithms needed for the extraction of the events that trigger both between and within sport camera selection, and describes a system that handles user preferences, live information and video-generated events to offer personalized content to the users.

2D and 3D face localization for complex scenes

In this paper, we address face tracking of multiple people in complex 3D scenes, using multiple calibrated and synchronized far-field recordings. We localize faces in every camera view and associate them across the different views. To cope with the complexity of 2D face localization introduced by the multitude of people and unconstrained face poses, a combination of stochastic and deterministic trackers, detectors and a Gaussian mixture model for face validation are utilized. Then faces of the same person seen from the different cameras are associated by first finding all possible associations and then choosing the best option by means of a 3D stochastic tracker. The performance of the proposed system is evaluated and is found enhanced compared to existing systems.

Person tracking for ambient camera selection in complex sports environments

As pervasive computing technologies are gradually penetrating sport, we are witnessing a proliferating number of research systems that can track athletes during training and/or in competition. Indeed, athlete tracking is a particularly challenging research tasks, which is also a key enabler for a wide range of applications such as ambient personalized broadcasting. In this paper we survey person tracking systems for sport applications and illustrate their limitations for realistic sports environments. As a real-life example we present the robust and high-performance person tracking system developed at the Athens Information Technology, and explain its inability to deal with occlusions, interlacing, adverse and changing light conditions and mostly the strain of the athletes, which are very common in high activity athletic scenes. We also present techniques for dealing with these challenges, along with an architecture for building ambient camera selection environments for broadcasting purposes. These techniques form the basis for the person tracking and ambient camera selection systems that are developed in the scope of the my-e-Director 2012 EC project, which is working towards a realistic prototype system for the London 2012 Olympics.

Tracking for context extraction in athletic events

Personalisation of large scale athletic events requires camera-specific annotations to provide for reasoning about incidents being best viewed by specific cameras. This needs an automatic system for annotating athletes on the video streams, to be achieved by the use of person trackers. In this paper we present a novel approach towards combining scene segmentation, motion outliers, face and bib tracks into body hypotheses and tracking them across time. Preliminary evaluation results demonstrate the potential of the proposed approach.

Face validation using 3D information from single calibrated camera

Detection of faces in cluttered scenes under arbitrary imaging conditions (pose, expression, illumination and distance) is prone to miss and false positive errors. The well-established approach of using boosted cascades of simple classifiers addresses the problem of missing faces by using fewer stages in the cascade. This constrains the misses by making detection easier, but increases the false positives. False positives can be reduced by validating the detected image regions as faces. This has been accomplished using color and pattern information of the detected image regions. In this paper we propose a novel face validation method based on 3D position estimates from a single calibrated camera. This is done by assuming a typical face width; hence the widths of the detected image regions lead to target position estimates. Detected image regions with extreme position estimates can then be discarded. We apply our method on the rich dataset of the CLEAR2007 evaluation campaign, comprising 49 thousand annotated indoors images, recorded at five different sites, from four different cameras per site, depicting approximately 122 thousand faces. Our method yields very accurate 3D position estimates, leading to superior results compared to color- and pattern-based face validation methods.

3D Tracking of Multiple People Using Their 2D Face Locations

In this paper, we address tracking of multiple people in complex 3D scenes, using multiple calibrated and synchronized far-field recordings. Our approach utilizes the faces detected in every camera view. Faces of the same person seen from the different cameras are associated by first finding all possible associations and then choosing the best option by means of a 3D stochastic tracker. The performance of the proposed system is evaluated by using the outputs of two grossly different 2D face detectors as input to our 3D algorithm. The multi-camera videos employed come from the CLEAR evaluation campaign. Even though the two 2D face detectors have very different performance, the 3D tracking performance of our system remains practically unchanged.