Publications

In this paper we address the problem of human posture classification, in particular focusing to an indoor surveillance application. The approach was initially inspired to a previous works of Haritaoglou et al. [6] that uses histogram projections to classify people’s posture. Projection histograms are here exploited as the main feature for the posture classification, but, differently from [6], we propose a supervised statistical learning phase to create probability maps adopted as posture templates. Moreover, camera calibration and homography is included to resolve prospective problems and improve the precision of classification. Furthermore, we make use of a finite state machineto detect dangerous situations as falls and to activate a suitable alarm generator. The system works on line on standard workstation with network cameras.

Typical processes of image representation comprehend initial region segmentation followed by a description of single regions’ feature and their relationships. Then a graph model can be exploited in order to integrate the knowledge of the specific regions (that are the attributed relational graph’s (ARG) nodes) and the regions’ relations (that are the ARG’s edges). In this work we use color features to guide region segmentation, geometric features to characterize regions one by one and topological features (and in particular inclusion) to describe regions’ relationships. Guided by the inclusion property we define the Topological Tree (TT) as an image representation model that exploiting the transitive property of inclusion, uses the adjacency and inclusion topological features. We propose an approach based on a recursive version of fuzzy c-means to construct the topological tree directly from the initial image, performing both segmentation and TT construction. The TT can be exploited in many applications of image analysis and image retrieval by similarity in those contexts where inclusion is a key feature: we propose an applicative case of analysis of dermatological images to support the melanoma diagnosis.In this paper describe details of the TT algorithm, including the management of not ideality and an approximate measure of tree similarity in order to retrieve skin lesion with a similar TT-based description.

In this work we present a framework for on-the-fly video transcoding that exploits computer vision-based techniques to adapt the Web access to the user requirements. Theproposed transcoding approach aims at coping with both user bandwidth and resources capabilities, and with user interests in the video's content. We propose an object-basedsemantic transcoding that, according to the user-dened classes of relevance, applies different transcoding techniques to the objects segmented in a scene. Object extraction is provided by on-the-fly video processing, without manual annotation. Multiple transcoding policies are reviewed and a performance evaluation metric based on the Weighted Mean Square Error (and corresponding PSNR), that takes into account the perceptual user requirements by means of classes of relevance, is dened. Results are analyzed by varying transcoding techniques, bandwidth requirements and video types (with indoor and outdoor scenes), showing that the use of semantics can dramatically improve the bandwidth to distortion ratio.

In contrast with common naevi, which generally show a homogeneous and regularly distributed pigmentation, brown to black pigment areas with irregular shape or asymmetric distribution are frequently observable in melanomas. Identification of dark areas inside a melanocytic lesion is of great importance for melanoma diagnosis, both during clinical examination and employing programs for automated image analysis. The aim of our study was to compare two different methods for the automated identification and description of dark areas in epiluminescence microscopy images of melanocytic lesions and to evaluate their diagnostic capability. 339 images of melanocytic lesions, referring to 113 melanomas and 226 melanocytic naevi, acquired by means of a polarizedlight videomicroscope (Videocap 200, DS-medica, Italy) with a 20 fold magnification were studied. Two different methods were employed for the identification of dark areas: the first permits the identification of ‘absolute’ dark areas, defined as areas which are darker than the skin. The second identifies the lesion area, the darkest with respect to the overall brightness of the lesion (‘relative’ dark areas). A set of parameters is extracted both for ‘absolute’ and ‘relative’ dark areas, in order to numerically describe the region properties, such as extension, balance, regularity and symmetry of its distribution. Significant differences in dark area distribution between melanomas and naevi were observed employing both methods, permitting a good discrimination of melanocytic lesions (diagnostic accuracy = 74.6% and 71.2% for absolute and relative dark areas, respectively). In conclusion, both methods for automated identification of dark areas are useful for melanoma diagnosis and can be implemented in programs for image analysis.

Monitoring and controlling the driver’s guidance by analyzing the rotation impressed to the steering-wheel can be a very important task in order to improve safety. This paper proposes a general-purpose method to track the steering wheel’s absolute angle by using a single camera vision system mounted inside the car. The absolute angle is computed by means of the accumulation of inter-frame relative rotations and the error propagation is prevented with an alignment process. The approach is based on the modeling of the motion of the steering wheel, as it appears perspectivelydistorted by the point of view of the un-calibrated camera. We modified the Lucas-Kanade method for an approximatively rotational motion model in order to provide the detection and tracking of significant features on the wheel. The experimental results are compared with ground-truthed data obtained with different types of sensors.

The paper presents an approach for a robust (semi-)automatic correction of radial lens distortion in images and videos. This method, based on the Hough transform, has the characteristics to be applicable also on videos from unknown cameras that, consequently, can not be a priori calibrated. We approximated the lens distortion by considering only the lower-order term of the radial distortion. Thus, the method relies on the assumption that pure radial distortion transforms straight lines into curves. The computation of the best value of the distortion parameter is performed in a multi-resolution way. The method precision depends on the scale of the multi-resolution and on the Hough space's resolution. Experiments are provided for both outdoor, uncalibrated camera and an indoor, calibrated one. The stability of the value found in different frames of the same video demonstrates the reliability of the proposed method.

In this paper we address the problem of fast segmenting moving objects in video acquired by moving camera or more generally with a moving background. We present an approach based on a color segmentation followed by a region-merging on motion through Markov Random Fields (MRFs). The technique we propose is inspired to a work of Gelgon and Bouthemy [6], that has been modified to reduce computational cost in order to achieve a fast segmentation (about ten frame per second). To this aim a modified region matching algorithm (namely Partitioned Region Matching) and an innovative arc-based MRF optimization algorithmwith a suitable definition of the motion reliability are proposed. Results on both synthetic and real sequences are reported to confirm validity of our solution.