Publications by Rita Cucchiara

Aim of the Visor Project [1] is to gather and makefreely available a repository of surveillance andvideo footages for the research community onpattern recognition and multimedia retrieval. Thegoal is to create an open forum and a free repositoryto exchange, compare and discuss results of manyproblems in video surveillance and retrieval.Together with the videos, the repository containsmetadata annotation, both manually annotated asground-truth and automatically obtained by videosurveillance systems. Annotation refers to a largeontology of concepts on surveillance and securityrelated objects and events. The ontology has beendefined including concepts from LSCOM andMediaMill ontologies. As well as videos andannotations, Visor provides tools for enriching theontology, annotating new videos, searching bytextual queries, composing and downloading videos.

In this paper we present a distributed videosurveillanceframework. Our end is the remote monitoringof the behavior of people moving in a scene exploitinga virtual reconstruction on low capabilitiesdevices, like PDAs and cell phones. The main noveltyof this system is the effective integration of the computervision and computer graphics modules. The first,using a probabilistic frameworks, can detect the position,the trajectory and the posture of peoples movingin the scene. The second exploits the new possibility ofboth standard 3D graphics libraries on mobile (namelyJSR184 and M3G graphic format) and new PDAsprocessing capability in order to reconstruct the remotesurveillance data in real-time.

Distributed video surveillance has a direct application in intelligent home automation or domotics (from the Latin word domus, that means “home”, and informatics); in particular, in-house videosurveillance can provide good support for people with some difficulties (e.g., elderly or disabled people) living alone and with a limited autonomy. New hardware technologies for surveillance are now affordable and provide high reliability. Problems related to reliable software solutions are not completely solved, especially concerning the application of general-purpose computer vision techniques in indoor environments. Indeed, assuming the objective is to detect the presence of people, track them, and recognize dangerous behaviours by means of abrupt changes in their posture, robust techniques must cope with non-trivial difficulties. In particular, luminance changes and shadows must be taken into account, frequent posture changes must be faced, and large and long-lasting occlusions are common due to the vicinity of the cameras and the presence of furnitureand doors that can often hide parts of the person’s body. These problems are analyzed and solutions based on background suppression, appearance-based probabilistic tracking, and probabilistic reasoning for posture recognition are described.

Pixel classifiers are often adopted in pattern recognition as a suitable method for image segmentation. A common approach to the performance evaluation of classifier systems is based on the measurement of the classification errors and, at the same time, on the computational time. In general, multiclassifiers have proven to be more precise in the classification in many applications, but at the cost of a higher computational load. This paper analyzes different classifiers and proposes an evaluation of the classifiers in the case of semi-automatic processes with human interaction. Medical imaging is a typical application, where automatic or semi-automatic segmentation can be a valuable support to the diagnosis. The paper focuses on the segmentation of cardiac images of fruit flies (genetic model for analyzing human heart's diseases). Analysis is based on M-modes, that are gray-level images derived from mono-dimensional projections of the video frames on a line. Segmentation of the M-mode images is provided by classifiers and integrated in a multiclassifier. A neural network classifier, a Bayesian classifier, and a classifier based on hidden Markov chains are joined by means of a Behavior Knowledge Space fusion rule. The comparative evaluation is discussed in terms of both accuracy and required time, in which the time to correct the classifier errors by means of human intervention is also taken into account.

In this work, we present a general purpose system for hierarchical structural segmentation and automatic annotation of video clips, by means of standardized low level features. We propose to automatically extract some prototypes for each class with a context based intra-class clustering. Clips are annotated following the MPEG-7 standard directives to provide easier portability. Results of automatic annotation and semiautomatic metadata creation are provided

This work proposes a method for automatic face obscuration capable of protecting people's identity. Since face detection heavily benefits from the possibility to exploit tracking, multi-camera people tracking has been integrated with a face detector based on colour clustering and Hough transform. Moreover, the multiple viewpoints provided by multiple cameras are exploited in order to always obtain a good-quality image of the face. The identity of people in different views is kept consistent by means of a geometrical, uncalibrated approach based on homographies. Experimental results show the accuracy of the proposed approach. (c) 2006 Elsevier B.V. All rights reserved.

In this paper an advanced video surveillance system is proposed.Our goal is the detection of the people’s heads toallow their obscuration for privacy issues or to performrecognition tasks. We propose a system based on active PTZ(Pan-Tilt-Zoom) cameras that produce head images havinga large enough size, and can cover an area larger than stillcameras. Since conventional approaches are not suitable toPTZ cameras, the proposed approach is based on the socalleddirection histograms to compute the ego-motion andon frame differencing for detecting moving objects. It exploitspost-processing and active contours to extract preciseshape of moving objects to be fed to a probabilistic algorithmto track moving people in the scene. Person following,instead, is based on simple heuristic rules that movethe camera as soon as the selected person is close to theborder of the field of view. Finally, a color and shape basedhead detection that takes advantage of the people trackingis presented. Experimental results on a live active camerademonstrate the feasibility of real-time person followingand of the consecutive head detection phase.

Automatic segmentation of skin lesions in clinical images is a very challenging task; it is necessary for visual analysis of the edges, shape and colors of the lesions to support the melanoma diagnosis, but, at the same time, it is cumbersome since lesions (both naevi and melanomas) do not have regular shape, uniform color, or univocal structure. Most of the approaches adopt unsupervised color clustering. This works compares the most spread color clustering algorithms, namely median cut, k-means, fuzzy-c means and mean shift applied to a method for automatic border extraction, providing an evaluation of the upper bound in accuracy that can be reached with these approaches. Different tests have been performed to examine the influence of the choice of the parameter settings with respect to the performances of the algorithms. Then a new supervised learning phase is proposed to select the best number of clusters and to segment the lesion automatically. Examples have been carried out in a large database of medical images, manually segmented by dermatologists. From these experiments mean shift was resulted the best technique, in term of sensitivity and specificity. Finally, a qualitative evaluation of the goodness of segmentation has been validated by the human experts too, confirming the results of the quantitative comparison.

In this paper we focus on the problem of automatically registering dermatological images, because even if different products are available, most of them share the problem of a limited field of view on the skin. A possible solution is then the composition of multiple takes of the same lesion with digital software, such as that for panorama images creation.In this work, to perform an automatic selection of matching points the Harris Corner Detector is used, and to cope with outlier couples we employed the RANSAC method. Projective mapping is then used to match the two images. Given a set of correspondence points, Singular Value Decomposition was used to compute the transform parameters.At this point the two images need to be blended together. One initial assumption is often implicitly made: the aim is to merge two rectangular images. But when merging occurs between more than two images iteratively, this assumption will fail. To cope with differently shaped images, we employed the Distance Transform and provided a weighted merging of images. Different tests were conducted with dermatological images, both with standard rectangular frame and with not typical shapes, as for example a ring due to the objective and lens selection. The successive composition of different circular images with other blending functions, such as the Hat function, doesn’t correctly get rid of the border and residuals of the circular mask are still visible. By applying Distance Transform blending, the result produced is insensitive of the outer shape of the image.

This paper proposes a novel method for estimating thegeospatial trajectory of a moving camera. The proposedmethod uses a set of reference images with known GPS(global positioning system) locations to recover the trajectoryof a moving camera using geometric constraints. Theproposed method has three main steps. First, scale invariantfeatures transform (SIFT) are detected and matched betweenthe reference images and the video frames to calculatea weighted adjacency matrix (WAM) based on the numberof SIFT matches. Second, using the estimated WAM, themaximum matching reference image is selected for the currentvideo frame, which is then used to estimate the relativeposition (rotation and translation) of the video frame usingthe fundamental matrix constraint. The relative position isrecovered upto a scale factor and a triangulation amongthe video frame and two reference images is performed toresolve the scale ambiguity. Third, an outlier rejection andtrajectory smoothing (using b-spline) post processing stepis employed. This is because the estimated camera locationsmay be noisy due to bad point correspondence or degenerateestimates of fundamental matrices. Results of recoveringcamera trajectory are reported for real sequences.