Gianpaolo Francesco Trotta

Computer vision and deep learning techniques for pedestrian detection and tracking: A survey

Abstract Pedestrian detection and tracking have become an important field in the computer vision research area. This growing interest, started in the last decades, might be explained by the multitude of potential applications that could use the results of this research field, e.g. robotics, entertainment, surveillance, care for the elderly and disabled, and content-based indexing. In this survey paper, vision-based pedestrian detection systems are analysed based on their field of application, acquisition technology, computer vision techniques and classification strategies. Three main application fields have been individuated and discussed: video surveillance, human-machine interaction and analysis. Due to the large variety of acquisition technologies, this paper discusses both the differences between 2D and 3D vision systems, and indoor and outdoor systems. The authors reserved a dedicated section for the analysis of the Deep Learning methodologies, including the Convolutional Neural Networks in pedestrian detection and tracking, considering their recent exploding adoption for such a kind systems. Finally, focusing on the classification point of view, different Machine Learning techniques have been analysed, basing the discussion on the classification performances on different benchmark datasets. The reported results highlight the importance of testing pedestrian detection systems on different datasets to evaluate the robustness of the computed groups of features used as input to classifiers.

A Novel Approach in Combination of 3D Gait Analysis Data for Aiding Clinical Decision-Making in Patients with Parkinson’s Disease

The most common methods used by neurologist to evaluate Parkinson’s Disease (PD) patients are rating scales, that are affected by subjective and non-repeatable observations. Since several research studies have revealed that walking is a sensitive indicator

A Supervised Breast Lesion Images Classification from Tomosynthesis Technique

In this paper, we propose a deep learning approach for breast lesions classification, by processing breast images obtained using an innovative acquisition system, the Tomosynthesis, a medical instrument able to acquire high-resolution images using a lower radiographic dose than normal Computed Tomography (CT). The acquired images were processed to obtain Regions Of Interest (ROIs) containing lesions of different categories. Subsequently, several pre-trained Convolutional Neural Network (CNN) models were evaluated as feature extractors and coupled with non-neural classifiers for discriminate among the different categories of lesions. Results showed that the use of CNNs as feature extractor and the subsequent classification using a non-neural classifier reaches high values of Accuracy, Sensitivity and Specificity.

Smell and Meaning: An OERP Study

The purpose of this work is to investigate the olfactory response to a neuter and a smell stimulation through Olfactory Event Related Potentials (OERP). We arranged an experiment of olfactory stimulation by analyzing Event Related Potential during perception of 2 odor stimuli: pleasant (Rose, 2-phenyl ethanol C2H4O2) and neuter (Neuter, Vaseline Oil CH2). We recruited 15 adult safe non-smokers volunteers. In order to record OERP, we used VOS EEG, a new device dedicated to odorous stimulation in EEG. After the OERP task, the subject filled a visual analogic scale, regarding the administered smell, on three dimensions: pleasantness (P), arousing (A) and familiarity (F). We performed an artificial neural network analysis that highlighted three groups of significant features, one for each amplitude component. Three neural network classifiers were evaluated in terms of accuracy on both full and restricted datasets, showing the best performance with the latter. The improvement of the accuracy rate in all VAS classifications was: 13.93% (A), 64.81% (F), 9.8% (P) for P300 amplitude (Fz); 16.28% (A), 49.46% (F), 24% (P) for N400 amplitude (Cz, Fz, O2, P8); 110.42% (A), 21.19% (F), 24.1% (P) for N600 amplitude (Cz, Fz). Main results suggested that in smell presentation we can observe the involvement of slow Event-Related-Potentials, like N400 and N600, ERP involved in stimulus encoding.

A model-free technique based on computer vision and sEMG for classification in Parkinson’s disease by using computer-assisted handwriting analysis

Abstract Patients suffering from Parkinson’s disease are characterized by an abnormal handwriting activity since they have difficulties in motor coordination and a decline in cognition. In this paper, we propose a model-free technique for differentiating Parkinson’s disease patients from healthy subjects by using a handwriting analysis tool based on computer vision and surface ElectroMyoGraphy (sEMG) signal-processing techniques and an Artificial Intelligence-based classifier. Experimental tests have been conducted with both healthy and Parkinson’s Disease patients using the proposed technique to address some specific research scientific questions regarding most representative features, best writing patterns, best AI-based classification approach between ANN optimal topology and SVM approaches in terms of both accuracy and repeatability of the results. Finally, the obtained results are reported and discussed to infer some important properties on writing patterns, classification approaches and the role of muscular activities on the handwriting analysis applied to neurodegenerative disease research.

Computer Vision and EMG-Based Handwriting Analysis for Classification in Parkinson's Disease.

Handwriting analysis represents an important research area in different fields. From forensic science to graphology, the automatic dynamic and static analyses of handwriting tasks allow researchers to attribute the paternity of a signature to a specific person or to infer medical and psychological patients’ conditions. An emerging research field for exploiting handwriting analysis results is the one related to Neurodegenerative Diseases (NDs). Patients suffering from a ND are characterized by an abnormal handwriting activity since they have difficulties in motor coordination and a decline in cognition.

A RGB-D Sensor Based Tool for Assessment and Rating of Movement Disorders

The assessment of tremor features of subjects affected by Parkinson’s disease supports physicians in defining customized rehabilitation treatment which, in turn, can lead to better clinical outcome. In the standard assessment protocol patient performed many exercises that are useful to physicians to rate disease. But the rating is subjective since is based on an observational evaluation. In this paper, we introduce a novel method for achieving objective assessment of movement conditions by directly measuring the magnitude of involuntary tremors with a set of sensors. We focused on one of the standard tasks of the Unified Parkinson’s Disease Rating Scale: finger-to-nose maneuver. During the task, data related to patient finger position are stored and then some tremor’s features are extracted. Finally, we employ a Support Vector Machine to measure the relevance of the extracted features in classify healthy subjects and patients.

A Wearable Device Supporting Multiple Touch- and Gesture-Based Languages for the Deaf-Blind

Over 1.5 million people in the world who are completely deaf-blind use touch-based alphabets to communicate with others and to interact with the world. However, they rely on an assistant who plays the role of an interpreter in translating the world for them. Unfortunately, despite the research work realized in the last decade, on the market there are no assistive devices for providing people who suffer from severe multi-sensory impairments with technology for social inclusion. In this paper, we introduce dbGLOVE, a wearable device for supporting deaf-blind people in being completely independent in communicating with others and in interacting with the world. Our system was co-designed with users to be a natural interface and to accommodate for different already-existing touch- and gesture-based languages, such as Malossi and deaf-blind manual, in order to offer a unique device for connecting different communities with an affordable solution.

Design of a Projective AR Workbench for Manual Working Stations

We present the design and a prototype of a projective AR workbench for an effective use of the AR in industrial applications, in particular for Manual Working Stations. The proposed solution consists of an aluminum structure that holds a projector and a camera that is intended to be mounted on manual working stations. The camera, using a tracking algorithm, computes in real time the position and orientation of the object while the projector displays the information always in the desired position. We also designed and implemented the data structure of a database for the managing of AR instructions, and we were able to access this information interactively from our application.

Recognition and Severity Rating of Parkinson’s Disease from Postural and Kinematic Features During Gait Analysis with Microsoft Kinect

When diagnosing Parkinson’s disease (PD), medical specialists normally assess several clinical manifestations of the patient and rate a severity level according to established criteria. They refer to the Movement Disorder Society – sponsored revision of Unified Parkinson’s Disease Rating Scale (MDS-UPDRS), the most widely adopted scale for rating PD. Since gait patterns differ between healthy elders and those with PD, we implement a simple, low-cost clinical tool that can extract kinematic and postural features through Microsoft Kinect v2 sensor to classify and rate PD. Thirty participants were enrolled for the purpose of the present study: sixteen PD patients, rated according to MDS-UPDRS Part IV for motor complications, and fourteen healthy paired subjects. Several gait cycles were extracted for each patient to improve the reliability of the methods and sixteen kinematic and postural features were considered. After preliminary feature selection, several classifier families were trained (both Support Vector Machine, SVM, and Artificial Neural Networks, ANN) and evaluated for the best solution. Results showed that the ANN classifier performed the best by reaching 89,40% of accuracy with only nine features in diagnosis PD and 95,02% of accuracy with only six features in rating PD severity.