Carla Maria Modena

A computer vision system for the detection and classification of vehicles at urban road intersections

The paper presents a real-time vision system to compute traffic parameters by analyzing monocular image sequences coming from pole-mounted video cameras at urban crossroads. The system uses a combination of segmentation and motion information to localize and track moving objects on the road plane, utilizing a robust background updating, and a feature-based tracking method. It is able to describe the path of each detected vehicle, to estimate its speed and to classify it into seven categories. The classification task relies on a model-based matching technique refined by a feature-based one for distinguishing between classes having similar models, like bicycles and motorcycles. The system is flexible with respect to the intersection geometry and the camera position. Experimental results demonstrate robust, real-time vehicle detection, tracking and classification over several hours of videos taken under different illumination conditions. The system is presently under trial in Trento, a 100,000-people town in northern Italy.

Automatic identification and skew estimation of text lines in real scene images

A method for the automatic localization of text embedded in complex images is proposed. It permits to detect the spatial position and the skew of the text lines which are present in the scene and to return a binary representation of each text line. Strengths of the algorithm are independence of text skew and of presence of connected text. After a preprocessing step the input image is segmented in order to obtain a set of connected components which represent the basic objects of the algorithm. Several heuristics are proposed to characterize text objects which depend both on the geometrical features of single components and on the geometrical and spatial relations among components. According to these heuristics several components are discarded and the retained ones are grouped into text lines candidates by means of a divisive hierachical clustering procedure. In the experimental session we describe the application of the algorithm to the extraction of text lines from the images of 100 book covers. Results about skew estimation are also reported.

Intelligent extended floating car data collection

The elaboration of data collected by vehicles moving on road network is relevant for traffic management and for private service providers, which can bundle updated traffic information with navigation services. Floating data, in its extended acceptation, contains not only time and location provided by a positioning system, but also information coming from various vehicle sensors. In this paper we describe our extended data collection system, in which vehicles are able to collect data about their local environment, namely the presence of roadworks and traffic slowdowns, by analyzing visual data taken by a looking forward camera and data from the on-board Electronic Control Unit. Upon detection of such events, a packet is set up containing time, position, vehicle data, results of on-board elaboration, one or more images of the road ahead and an estimation of the local traffic level. Otherwise, the transmitted packet containing only the minimal data, making its size adaptive to the environment surrounding the vehicle.

A kalman filter based background updating algorithm robust to sharp illumination changes

A novel algorithm, based on Kalman filtering is presented for updating the background image within video sequences. Unlike existing implementations of the Kalman filter for this task, our algorithm is able to deal with both gradual and sudden global illumination changes. The basic idea is to measure global illumination change and to use it as an external control of the filter. This allows the system to better fit the assumptions about the process to be modeled. Moreover, we propose methods to estimate measurement noise variance and to deal with the problem of saturated pixels, to improve the accuracy and robustness of the algorithm. The algorithm has been successfully tested in a traffic surveillance task by comparing it to a background updating algorithm, based on Kalman filtering, taken from literature.

A smart watch-based gesture recognition system for assisting people with visual impairments

Modern mobile devices provide several functionalities and new ones are being added at a breakneck pace. Unfortunately browsing the menu and accessing the functions of a mobile phone is not a trivial task for visual impaired users. Low vision people typically rely on screen readers and voice commands. However, depending on the situations, screen readers are not ideal because blind people may need their hearing for safety, and automatic recognition of voice commands is challenging in noisy environments. Novel smart watches technologies provides an interesting opportunity to design new forms of user interaction with mobile phones. We present our first works towards the realization of a system, based on the combination of a mobile phone and a smart watch for gesture control, for assisting low vision people during daily life activities. More specifically we propose a novel approach for gesture recognition which is based on global alignment kernels and is shown to be effective in the challenging scenario of user independent recognition. This method is used to build a gesture-based user interaction module and is embedded into a system targeted to visually impaired which will also integrate several other modules. We present two of them: one for identifying wet floor signs, the other for automatic recognition of predefined logos.

An efficient vehicle queue detection system based on image processing

This paper describes a method for the real-time measurement of vehicle queue parameters in a video-based traffic monitoring experimental system. The method proposed here is based on vehicle presence detection and movement analysis in video sequences acquired by a stationary camera. Queues are detected and characterized by a severity index. Intensive experiments show the robustness of the method under varying illumination and weather conditions. The system is presently undergoing an on-field testing phase in a double ways road near Trento, Italy, where queues frequently occur.

Vision-based bicycle/motorcycle classification

We present a feature-based classifier that distinguishes bicycles from motorcycles in real-world traffic scenes. The algorithm extracts some visual features focusing on the wheel regions of the vehicles. It splits the problem into two sub-cases depending on the computed motion direction. The classification is performed by non-linear Support Vector Machines. Tests lead to a successful vehicle classification rate of 96.7% on video sequences taken from different road junctions in an urban environment.

Scene text recognition and tracking to identify athletes in sport videos

We present an athlete identification module forming part of a system for the personalization of sport video broadcasts. The aim of this module is the localization of athletes in the scene, their identification through the reading of names or numbers printed on their uniforms, and the labelling of frames where athletes are visible. Building upon a previously published algorithm we extract text from individual frames and read these candidates by means of an optical character recognizer (OCR). The OCR-ed text is then compared to a known list of athletes’ names (or numbers), to provide a presence score for each athlete. Text regions are tracked in subsequent frames using a template matching technique. In this way blurred or distorted text, normally unreadable by the OCR, is exploited to provide a denser labelling of the video sequences. Extensive experiments show that the method proposed is fast, robust and reliable, out-performing results of other systems in the literature.

Personal Shopping Assistance and Navigator System for Visually Impaired People

In this paper, a personal assistant and navigator system for visually impaired people will be described. The showcase presented intends to demonstrate how partially sighted people could be aided by the technology in performing an ordinary activity, like going to a mall and moving inside it to find a specific product. We propose an Android application that integrates Pedestrian Dead Reckoning and Computer Vision algorithms, using an off-the-shelf Smartphone connected to a Smartwatch. The detection, recognition and pose estimation of specific objects or features in the scene derive an estimate of user location with sub-meter accuracy when combined with a hardware-sensor pedometer. The proposed prototype interfaces with a user by means of Augmented Reality, exploring a variety of sensorial modalities other than just visual overlay, namely audio and haptic modalities, to create a seamless immersive user experience. The interface and interaction of the preliminary platform have been studied through specific evaluation methods. The feedback gathered will be taken into consideration to further improve the proposed system.

Boosting Fisher vector based scoring functions for person re-identification

In recent years, much effort has been put into the development of novel algorithms to solve the person re-identification problem. The goal is to match a given persons image against a gallery of people. In this paper, we propose a single-shot supervised method to compute a scoring function that, when applied to a pair of images, provides a score expressing the likelihood that they depict the same individual. The method is characterized by: (i) the usage of a set of local image descriptors based on Fisher vectors, (ii) the training of a pool of scoring functions based on the local descriptors, and (iii) the construction of a strong scoring function by means of an adaptive boosting procedure. The method has been tested on four data-sets and results have been compared with state-of-the-art methods clearly showing superior performance. Display Omitted We propose BFiVe, a new supervised algorithm for single-shot person re-identification.The descriptors are a set of compressed local Fisher vectors extracted from a coarse to fine image subdivision.In the training step each region gives rise to a learnt weak ranking function.The ranking function of the image gallery is obtained by a boosted selection of a weak learner subset.The matching rate at rank 1 on VIPeR is 38.9%, on 3DPes 41.7%, on PRID-2011 19.6%, and on i-LIDS-119 48.1%.