Search

The Purpose of this project is the control of motion and direction in real time of a wheel chair, using machine vision algorithms. The main goal of this project is signal acquisition from the video camera and collision sensors for post processing in the C# algorithms and later obtaining motor control in the traction mechanism of the wheelchair. The C# algorithm has several tasks. The first is to obtain the real time image from web cam and later processing for the identification of the direction of movement of the human face. The second is to calculate the speed of the movement for generation of the PWM output for motor movement. This information output uses the RS232C driver with a microcontroller card attached to a motor control box in the wheel chair mechanism. The final task is to obtain the collision sensor status for security implementation, all in real time. The main reason for development of an implementation of this solution is the use of open source software tools for a more stable platform in the base system due to the characteristics of the end use of the system. The end user of the system will be a quadriplegic.

This article proposes and implements a network covert channel called InCC capable of hiding information on the Internet, which is designed to produce a undetectable communication channel between systems. This network channel is fully transparent to any network analysis and for hence to any interception and inspection on a network. InCC is capable to send messages on the same production network without compromising the existence of source and destination. By using techniques like encryption, address spoofing, signature poisoning and traffic analysis, the channel is able to hide the flows on the network without implicating the source and destination.

A novel methodology for person identification based on soft biometrics and oriented to applications in real surveillance environments is proposed in this paper. For this purpose, a visual analysis of a bag-of-soft-biometrics features related to color, texture, local features and geometry extracted from individuals is carried out. To measure the relevance of each extracted feature, several methods are proposed. Moreover, features are weighted in two different ways according to their importance. Each method is evaluated under two different scenarios: mono-camera and multi-camera images. With the aim of testing the system in a realistic way, it has been evaluated over standard databases in the surveillance community: PETS 2006, PETS 2009 and CAVIAR. An analysis of relevance for each of features acquired in these two scenarios is presented.

This paper presents and an evaluation of results obtained from a face recognition system in a real uncontrolled localization. The involved infrastructure is Barajas Airport (the international airport in Madrid, Spain). The use of this infrastructure during normal operation hours has imposed some constrains. It was not allowed to change or to add new cameras and passengers should not be disturbed by any means. Passengers should not be aware of the presence of the system, so no request should be done to change their normal behavior. To fulfill these requirements, three video surveillance cameras were selected: two in the corridor areas and one in a control point. Images were acquired and processed with illumination changes, several quality levels, collaborative and non-collaborative subjects and during three weeks. The influence of data compression method and classificator has been detailed in the paper. Three scenarios were simulated: first one is a normal operational mode, second one is a high security mode and last one is a friendly or soft-recognition mode. Four data compression methods were considered in the paper: 1dpca (1d principal components analysis), 2dpca (2d principal components analysis), 2dlda (2d linear discriminant analysis) and csa (coupled subspace analysis). Csa has obtained the best performance. For classificatory purpoises, svm (support vector machines) were selected with excellent results. The overall analysis shows that the approach taken will lead to excellent results given the hard conditions of a real scenario such an airport.

The aim of this paper is to present a novelty methodology to develop similarity measures for classification of time series. First, a linear segmentation algorithm to obtain a section-wise representation of the series is presented. Then, two similarity measures are defined from the differences between the behavior of the series and the level of the series, respectively. The method is applied to subjective-data on time series generated through the evaluations of the driving risk from a group of traffic safety experts. These series are classified using the proposed similarities as kernels for the training of a Support Vector Machine. The results are compared with other classifiers using our similarities, their linear combination and the raw data. The proposed methodology has been successfully evaluated on several databases.

In this paper a complete driver's hands detection and tracking system suitable for working in real time conditions has been developed. The proposed system has been successfully tested in close-real world conditions in different scenarios on a very realistic and immersive cabin truck simulator. A database of 24 video sequences monitoring the driving task in different circuits, illumination conditions and video resolutions has been obtained. The hands detection rate and the computational times needed to process each frame are presented. The proposed system has proven to be high accurate and fast enough to work in real time conditions. In the future, the selected algorithm will be included as part of an automotive compliance embedded system placed in a real truck cabin.

We present a face verification system using Parallel Gabor Principal Component Analysis (PGPCA) and fusion of Support Vector Machines (SVM) scores. The algorithm has been tested on two databases: XM2VTS (frontal images with frontal or lateral illumination) and FRAV2D (frontal images with diffuse or zenithal illumination, varying poses and occlusions). Our method outperforms others when fewer PCA coefficients are kept. It also has the lowest equal error rate (EER) in experiments using frontal images with occlusions. We have also studied the influence of wavelet frequency and orientation on the EER in a one-Gabor PCA. The high frequency wavelets are able to extract more discriminant information compared to the low frequency wavelets. Moreover, as a general rule, oblique wavelets produce a lower EER compared to horizontal or vertical wavelets. Results also suggest that the optimal wavelet orientation coincides with the illumination gradient.

The problem of high dimensionality in face verification tasks has recently been simplified by the use of underlying spatial structures as proposed in the Two Dimensional Principal Component Analysis, the Two Dimensional Linear Discriminant Analysis and the Coupled Subspaces Analysis. Besides, the Small Sample Size problem that caused serious difficulties in traditional LDA has been overcome by the spatial approach 2DLDA. The application of these advances to facial verification techniques using different SVM schemes as classification algorithm is here shown. The experiments have been performed over a wide facial database (FRAV2D including 109 subjects), in which only one interest variable was changed in each experiment: illumination, pose, expression or occlusion. For training the SVMs, only two images per subject have been provided to fit in the small sample size problem.

The problem of high dimensionality in face verification tasks has recently been simplified by the use of underlying spatial structures as proposed in the 2DPCA, 2DLDA and CSA methods. Fusion techniques at both levels, feature extraction and matching score, have been developed to join the information obtained and achieve better results in verification process. The application of these advances to facial verification techniques using different SVM schemes as classification algorithm is here shown. The experiments have been performed over a wide facial database (FRAV2D including 109 subjects), in which only one interest variable was changed in each experiment. For training the SVMs, only two images per subject have been provided to fit in the small sample size problem.

We present a novel 3D facial feature location method based on the Spin Images registration technique. Three feature points are localized: the nose tip and the inner corners of the right and left eye. The points are found directly in the 3D mesh, allowing a previous normalization before the depth map calculation. This method is applied after a preprocess stage where the candidate points are selected measuring curvatures on the surface and applying clustering techniques. The system is tested on a 3D Face Database called FRAV3D with 105 people and a widely variety of acquisition conditions in order to test the method in a non-controlled environment. The success location rate is 99.5% in the case of the nose tip and 98% in the case of eyes, in frontal conditions. This rate is similar even if the conditions change allowing small rotations. Results in more extremely acquisition conditions are shown too. A complete study of the influence of the mesh resolution over the spin images quality and therefore over the face feature location rate is presented. The causes of the errors are discussed in detail.

In this paper two methods for human face recognition and the influence of location mistakes are shown. First one, Principal Components Analysis (PCA), has been one of the most applied methods to perform face verification in 2D. In our experiments three classifiers have been considered to test influence of location errors in face verification using PCA. An initial set of ”correct located faces” has been used for PCA matrix computation and to train all classifiers. An initial test set was built considering a ”correct located faces” set (based on different images than training ones) and then a new test set was obtained by applying a small displacement in both axis (20 pixels) to the initial set. Second method is based on geometrical characteristics constructed with facial and cranial points that come from a 3D representation. Data are acquired by a calibrated stereo system. Classifiers considered for both methods are k-nearest neighbours (KNN), artificial neural networks: radial basis function (RBF) and Support Vector Machine (SVM). Given our data set, results show that SVM is capable to classify correctly in the presence of small location errors. RBF has an acceptable correct rate but the number of false positives is always higher than in the SVM case.

The design, development, construction and testing of an Artificial-Vision controlled Traffic-Light prototype has been carried out to rule and regulate intersections. Methods, algorithms and automatons have been built up with that purpose to provide the analysis of images and decisions making at real time. The aim has been the development of an intelligent traffic-light capable of capturing the presence or absence of vehicles, pedestrians and their particular situations defined by their trajectories. Besides the above mentioned properties we have to point out the adaptation to the precise characteristics of each crossing, as its geometry, the required equipment, etc. The project has been supervised by RACE, world wide known as experts in road safety awareness, endowing the prototype with reliability and trust.

Spatial dimension reduction methods called Two Dimensional PCA and Two Dimensional LDA have recently been presented. These variations of traditional PCA and LDA consider images as 2D matrices instead of 1D vectors. The robustness to pose variations of these advances at verification tasks, using SVM as classification algorithm, is here shown. The new methods endowed with a classification strategy of SVMs, seriously improve, specially for pose variations, the results achieved by the traditional classification of PCA and SVM.

Spatial dimension reduction called Two Dimensional PCA method has recently been presented. The application of this variation of traditional PCA considers images as 2D matrices instead of 1D vectors as other dimension reduction methods have been using. The application of these advances to verification techniques, using SVM as classification algorithm, is here shown. The simulation has been performed over a complete facial images database called FRAV2D that contains different sets of images to measure the improvements on several difficulties such as rotations, illumination problems, gestures or occlusion. The new method endowed with a classification strategy of SVMs, seriously improves the results achieved by the traditional classification of PCA & SVM.

Influence of illumination conditions in face verification using support vector machines (SVM) and k-nearest neighbours is analysed using an experimental set up in which images are acquired in controlled or uncontrolled illumination conditions. Principal components analysis (PCA) has been considered to perform dimensional reduction. SVM techniques offers better results even if linear kernels are considered.

We present a face verification system with an acceptance threshold automatically computed. The user is allowed to provide the rate between the costs assumed for a false acceptance and false rejection. This rate between costs can be intuitively known by the system responsible and are a starting point to fulfil user security requirements. With this user-friendly data, an algorithm based on screening techniques to compute the acceptance threshold is presented in this paper. This algorithm is applied to an original and competitive face verification system based on principal component analysis and two classifiers (neural network radial basis function and support vector machine). Experimental results with a 100 people face database are shown. This method can be also applied into other biometric applications in which this threshold should be calculated.

This paper presents some results on the possibilities offered by neural networks for human face recognition. In particular, two algorithms have been tested: learning vector quantization (LVQ) and multilayer perception (MLP). Two different approaches have been taken for each case, using as input data either preprocessed images (gray level or segmented), or geometrical features derived from a set of manually introduced landmarks. The preprocessing steps included resolution reduction and segmentation. For the geometrical features’ case, a Karhunen-Loeve expansion was used to extract features among the different possibilities offered by 14 landmark points.