Walter Nunziati

Soccer highlights detection and recognition using HMMs

In this paper we report on our experience in the detection and recognition of soccer highlights in videos using hidden Markov models. A first approach relies on camera motion only, whereas a second one also includes information regarding the location of players on the playing field. While the former approach requires less information, the latter has proven to be more precise. Our experimental evaluation yields interesting results.

Video clip matching using MPEG-7 descriptors and edit distance

Video databases require that clips are represented in a compact and discriminative way, in order to perform efficient matching and retrieval of documents of interest. We present a method to obtain a video representation suitable for this task, and show how to use this representation in a matching scheme. In contrast with existing works, the proposed approach is entirely based on features and descriptors taken from the well established MPEG-7 standard. Different clips are compared using an edit distance, in order to obtain high similarity between videos that differ for some subsequences, but are essentially related to the same content. Experimental validation is performed using a prototype application that retrieves TV commercials recorded from different TV sources in real time. Results show excellent performances both in terms of accuracy, and in terms of computational performances.

Improving evidential quality of surveillance imagery through active face tracking

We report on a system for automatically obtaining high-resolution images of surveillance targets. The system, beginning from a framed target, uses the presence of faces as a cue to initiate a real-time active face tracking process. Using a single pan-tilt-zoom (PTZ) camera, the system tracks the target while zooming in for a closeup image of the face. The system tracks feature points on a target face, and a simple yet reliable control strategy is used in conjunction with robust target localization to guide a PTZ camera to a facial closeup of the target. We report on experiments with our system in real surveillance environments

Automatic extraction and annotation of soccer video highlights

Broadcasters are demonstrating interest in systems that ease the process of annotation the huge amount of live and archived video materials. Exploitation of such assets is considered a key method for the improvement of production quality, and sport videos are one of the most marketable assets. In particular, in Europe, soccer is one of the most relevant sport types. This paper deals with detection and recognition of soccer highlights, using an approach based on temporal logic models.

Soccer players identification based on visual local features

Semantic detection and recognition of objects and events contained in a video stream has to be performed in order to provide content-based annotation and retrieval of videos. This annotation is done as a means to be able to reuse the video material at a later stage, e.g. to produce new TV programmes. A typical example is that of sports videos, where videos are annotated in order to reuse the video clips that show key highlights and players to produce short summaries for news and sports programmes. In order to select the most interesting actions among all the possibly detected highlights further analysis is required; i.e. the shots that contain a key action are typically followed by close-ups of the players that take part in the action. Therefore the automatic identification of these players would add considerable value both to the annotation and retrieval of the key highlights and key players of a sport event. The problem of detecting and recognizing faces in broadcast videos is a widely studied topic. However, in the case of soccer videos, and sports videos in general, the current techniques are not suitable for the task of face recognition, due to the high variations in pose, illumination, scale and occlusion that may happen in an uncontrolled environment. In this paper a method that copes with these problems, exploiting local features to describe a face, without requiring a precise localization of the distinguishing parts of a face, and the set of poses to describe a person and perform a more robust recognition, is presented. A similarity metric based on the number of matched interest points, able to cope with different face sizes, is also presented and experimentally validated.

Player identification in soccer videos

A method for the identification of players in soccer videos is presented. The proposed approach exploits the inherent multiple media structure of soccer videos to perform people identification without relying on face recognition. Instead, faces are detected in close-up shots, and then the filmed player is recognized by means of recognition of the number depicted on the frontal part of its jersey, or by detection and interpretation of superimposed closed caption. Players not identified by this process are then assigned to one of the labeled faces by means of a face similarity measure. We present results obtained from soccer videos of the last European Championship for national teams, held in Portugal in June 2004.

Automatic interpretation of soccer video for highlights extraction and annotation

Broadcasters are demonstrating interest in systems that ease the process of annotation the huge amount of live and archived video materials. Exploitation of such assets is considered a key method for the improvement of production quality, and sport videos are one of the most marketable assets. In particular, in Europe, soccer is one of the most relevant sport types. This paper deals with detection and recognition of soccer highlights, using an approach based on temporal logic models.

Adaptive uncertainty estimation for particle filter-based trackers

In particle filter-based visual trackers, dynamic velocity components are typically incorporated into the state update equations. In these cases, there is a risk that the uncertainty in the model update stage can become amplified in unexpected and undesirable ways, leading to erroneous behavior of the tracker. To deal with this problem, we propose a continuously adaptive approach to estimating uncertainty in the particle filter, one that balances the uncertainty in its static and dynamic elements. We provide quantitative performance evaluation of the resulting particle filter tracker on a set of ten video sequences. Results are reported in terms of a metric that can be used to objectively evaluate the performance of visual trackers. This metric is used to compare our modified particle filter tracker and the continuously adaptive mean shift tracker. Results show that the performance of the particle filter is significantly improved through adaptive parameter estimation, particularly in cases of occlusions and nonlinear target motion.

Highlights modeling and detection in sports videos

Automatic annotation of semantic events allows effective retrieval of video content. In this work, we present solutions for highlights detection in sports videos. This application is particularly interesting for broadcasters, since they extensively use manual annotation to select interesting highlights that are edited to create new programmes. The proposed approach exploits the typical structure of a wide class of sports videos, namely, those related to sports which are played in delimited venues with playfields of well known geometry, like soccer, basketball, swimming, track and field disciplines, and so on. For this class of sports, a modeling scheme based on a limited set of visual cues and on finite state machines (FSM) that encode the temporal evolution of highlights is presented. Algorithms for model checking and for visual cues estimation are discussed, as well as applications of the representation to different sport domains.

Improving the robustness of particle filter-based visual trackers using online parameter adaptation

In particle filter-based visual trackers, dynamic velocity components are typically incorporated into the state update equations. In these cases, there is a risk that the uncertainty in the model update stage can become amplified in unexpected and undesirable ways, leading to erroneous behavior of the tracker. Moreover, the use of a weak appearance model can make the estimates provided by the particle filter inaccurate. To deal with this problem, we propose a continuously adaptive approach to estimating uncertainty in the particle filter, one that balances the uncertainty in its static and dynamic elements. We provide quantitative performance evaluation of the resulting particle filter tracker on a set of ten video sequences. Results are reported in terms of a metric that can be used to objectively evaluate the performance of visual trackers. This metric is used to compare our modified particle filter tracker and the continuously adaptive mean shift tracker. Results show that the performance of the particle filter is significantly improved through adaptive parameter estimation, particularly in cases of occlusion and erratic, nonlinear target motion.