Catadioptric panoramic stereovision for humanoid robots

This paper proposes a novel design of a reconfigurable humanoid robot head, based on biological likeness of human being so that the humanoid robot could agreeably interact with people in various everyday tasks. The proposed humanoid head has a modular and adaptive structural design and is equipped with three main components: frame, neck motion system and omnidirectional stereovision system modules. The omnidirectional stereovision system module being the last module, a motivating contribution with regard to other computer vision systems implemented in former humanoids, it opens new research possibilities for achieving human-like behaviour. A proposal for a real-time catadioptric stereovision system is presented, including stereo geometry for rectifying the system configuration and depth estimation. The methodology for an initial approach for visual servoing tasks is divided into two phases, first related to the robust detection of moving objects, their depth estimation and position calculation, and second the development of attention-based control strategies. Perception capabilities provided allow the extraction of 3D information from a wide range of visions from uncontrolled dynamic environments, and work results are illustrated through a number of experiments.

Real-time image processing for crop/weed discrimination in maize fields

This paper presents a computer vision system that successfully discriminates between weed patches and crop rows under uncontrolled lighting in real-time. The system consists of two independent subsystems, a fast image processing delivering results in real-time (Fast Image Processing, FIP), and a slower and more accurate processing (Robust Crop Row Detection, RCRD) that is used to correct the first subsystem's mistakes. This combination produces a system that achieves very good results under a wide variety of conditions. Tested on several maize videos taken of different fields and during different years, the system successfully detects an average of 95% of weeds and 80% of crops under different illumination, soil humidity and weed/crop growth conditions. Moreover, the system has been shown to produce acceptable results even under very difficult conditions, such as in the presence of dramatic sowing errors or abrupt camera movements. The computer vision system has been developed for integration into a treatment system because the ideal setup for any weed sprayer system would include a tool that could provide information on the weeds and crops present at each point in real-time, while the tractor mounting the spraying bar is moving

Reconocimiento facial

El presente proyecto trata sobre uno de los campos más problemáticos de la inteligencia artificial, el reconocimiento facial. Algo tan sencillo para las personas como es reconocer una cara conocida se traduce en complejos algoritmos y miles de datos procesados en cuestión de segundos. El proyecto comienza con un estudio del estado del arte de las diversas técnicas de reconocimiento facial, desde las más utilizadas y probadas como el PCA y el LDA, hasta técnicas experimentales que utilizan imágenes térmicas en lugar de las clásicas con luz visible. A continuación, se ha implementado una aplicación en lenguaje C++ que sea capaz de reconocer a personas almacenadas en su base de datos leyendo directamente imágenes desde una webcam. Para realizar la aplicación, se ha utilizado una de las librerías más extendidas en cuanto a procesado de imágenes y visión artificial, OpenCV. Como IDE se ha escogido Visual Studio 2010, que cuenta con una versión gratuita para estudiantes. La técnica escogida para implementar la aplicación es la del PCA ya que es una técnica básica en el reconocimiento facial, y además sirve de base para soluciones mucho más complejas. Se han estudiado los fundamentos matemáticos de la técnica para entender cómo procesa la información y en qué se datos se basa para realizar el reconocimiento. Por último, se ha implementado un algoritmo de testeo para poder conocer la fiabilidad de la aplicación con varias bases de datos de imágenes faciales. De esta forma, se puede comprobar los puntos fuertes y débiles del PCA. ABSTRACT. This project deals with one of the most problematic areas of artificial intelligence, facial recognition. Something so simple for human as to recognize a familiar face becomes into complex algorithms and thousands of data processed in seconds. The project begins with a study of the state of the art of various face recognition techniques, from the most used and tested as PCA and LDA, to experimental techniques that use thermal images instead of the classic visible light images. Next, an application has been implemented in C + + language that is able to recognize people stored in a database reading images directly from a webcam. To make the application, it has used one of the most outstretched libraries in terms of image processing and computer vision, OpenCV. Visual Studio 2010 has been chosen as the IDE, which has a free student version. The technique chosen to implement the software is the PCA because it is a basic technique in face recognition, and also provides a basis for more complex solutions. The mathematical foundations of the technique have been studied to understand how it processes the information and which data are used to do the recognition. Finally, an algorithm for testing has been implemented to know the reliability of the application with multiple databases of facial images. In this way, the strengths and weaknesses of the PCA can be checked.

Correction of the Colour Variations under Uncontrolled Lighting Conditions

Esta tesis trata sobre métodos de corrección que compensan la variación de las condiciones de iluminación en aplicaciones de imagen y video a color. Estas variaciones hacen que a menudo fallen aquellos algoritmos de visión artificial que utilizan características de color para describir los objetos. Se formulan tres preguntas de investigación que definen el marco de trabajo de esta tesis. La primera cuestión aborda las similitudes que se dan entre las imágenes de superficies adyacentes en relación a su comportamiento fotométrico. En base al análisis del modelo de formación de imágenes en situaciones dinámicas, esta tesis propone un modelo capaz de predecir las variaciones de color de la región de una determinada imagen a partir de las variaciones de las regiones colindantes. Dicho modelo se denomina Quotient Relational Model of Regions. Este modelo es válido cuando: las fuentes de luz iluminan todas las superficies incluídas en él; estas superficies están próximas entre sí y tienen orientaciones similares; y cuando son en su mayoría lambertianas. Bajo ciertas circunstancias, la respuesta fotométrica de una región se puede relacionar con el resto mediante una combinación lineal. No se ha podido encontrar en la literatura científica ningún trabajo previo que proponga este tipo de modelo relacional. La segunda cuestión va un paso más allá y se pregunta si estas similitudes se pueden utilizar para corregir variaciones fotométricas desconocidas en una región también desconocida, a partir de regiones conocidas adyacentes. Para ello, se propone un método llamado Linear Correction Mapping capaz de dar una respuesta afirmativa a esta cuestión bajo las circunstancias caracterizadas previamente. Para calcular los parámetros del modelo se requiere una etapa de entrenamiento previo. El método, que inicialmente funciona para una sola cámara, se amplía para funcionar en arquitecturas con varias cámaras sin solape entre sus campos visuales. Para ello, tan solo se necesitan varias muestras de imágenes del mismo objeto capturadas por todas las cámaras. Además, este método tiene en cuenta tanto las variaciones de iluminación, como los cambios en los parámetros de exposición de las cámaras. Todos los métodos de corrección de imagen fallan cuando la imagen del objeto que tiene que ser corregido está sobreexpuesta o cuando su relación señal a ruido es muy baja. Así, la tercera cuestión se refiere a si se puede establecer un proceso de control de la adquisición que permita obtener una exposición óptima cuando las condiciones de iluminación no están controladas. De este modo, se propone un método denominado Camera Exposure Control capaz de mantener una exposición adecuada siempre y cuando las variaciones de iluminación puedan recogerse dentro del margen dinámico de la cámara. Los métodos propuestos se evaluaron individualmente. La metodología llevada a cabo en los experimentos consistió en, primero, seleccionar algunos escenarios que cubrieran situaciones representativas donde los métodos fueran válidos teóricamente. El Linear Correction Mapping fue validado en tres aplicaciones de re-identificación de objetos (vehículos, caras y personas) que utilizaban como caracterísiticas la distribución de color de éstos. Por otra parte, el Camera Exposure Control se probó en un parking al aire libre. Además de esto, se definieron varios indicadores que permitieron comparar objetivamente los resultados de los métodos propuestos con otros métodos relevantes de corrección y auto exposición referidos en el estado del arte. Los resultados de la evaluación demostraron que los métodos propuestos mejoran los métodos comparados en la mayoría de las situaciones. Basándose en los resultados obtenidos, se puede decir que las respuestas a las preguntas de investigación planteadas son afirmativas, aunque en circunstancias limitadas. Esto quiere decir que, las hipótesis planteadas respecto a la predicción, la corrección basada en ésta y la auto exposición, son factibles en aquellas situaciones identificadas a lo largo de la tesis pero que, sin embargo, no se puede garantizar que se cumplan de manera general. Por otra parte, se señalan como trabajo de investigación futuro algunas cuestiones nuevas y retos científicos que aparecen a partir del trabajo presentado en esta tesis. ABSTRACT This thesis discusses the correction methods used to compensate the variation of lighting conditions in colour image and video applications. These variations are such that Computer Vision algorithms that use colour features to describe objects mostly fail. Three research questions are formulated that define the framework of the thesis. The first question addresses the similarities of the photometric behaviour between images of dissimilar adjacent surfaces. Based on the analysis of the image formation model in dynamic situations, this thesis proposes a model that predicts the colour variations of the region of an image from the variations of the surrounded regions. This proposed model is called the Quotient Relational Model of Regions. This model is valid when the light sources illuminate all of the surfaces included in the model; these surfaces are placed close each other, have similar orientations, and are primarily Lambertian. Under certain circumstances, a linear combination is established between the photometric responses of the regions. Previous work that proposed such a relational model was not found in the scientific literature. The second question examines whether those similarities could be used to correct the unknown photometric variations in an unknown region from the known adjacent regions. A method is proposed, called Linear Correction Mapping, which is capable of providing an affirmative answer under the circumstances previously characterised. A training stage is required to determine the parameters of the model. The method for single camera scenarios is extended to cover non-overlapping multi-camera architectures. To this extent, only several image samples of the same object acquired by all of the cameras are required. Furthermore, both the light variations and the changes in the camera exposure settings are covered by correction mapping. Every image correction method is unsuccessful when the image of the object to be corrected is overexposed or the signal-to-noise ratio is very low. Thus, the third question refers to the control of the acquisition process to obtain an optimal exposure in uncontrolled light conditions. A Camera Exposure Control method is proposed that is capable of holding a suitable exposure provided that the light variations can be collected within the dynamic range of the camera. Each one of the proposed methods was evaluated individually. The methodology of the experiments consisted of first selecting some scenarios that cover the representative situations for which the methods are theoretically valid. Linear Correction Mapping was validated using three object re-identification applications (vehicles, faces and persons) based on the object colour distributions. Camera Exposure Control was proved in an outdoor parking scenario. In addition, several performance indicators were defined to objectively compare the results with other relevant state of the art correction and auto-exposure methods. The results of the evaluation demonstrated that the proposed methods outperform the compared ones in the most situations. Based on the obtained results, the answers to the above-described research questions are affirmative in limited circumstances, that is, the hypothesis of the forecasting, the correction based on it, and the auto exposure are feasible in the situations identified in the thesis, although they cannot be guaranteed in general. Furthermore, the presented work raises new questions and scientific challenges, which are highlighted as future research work.

Multi-sensor background subtraction by fusing multiple region-based probabilistic classifiers

In the recent years, the computer vision community has shown great interest on depth-based applications thanks to the performance and flexibility of the new generation of RGB-D imagery. In this paper, we present an efficient background subtraction algorithm based on the fusion of multiple region-based classifiers that processes depth and color data provided by RGB-D cameras. Foreground objects are detected by combining a region-based foreground prediction (based on depth data) with different background models (based on a Mixture of Gaussian algorithm) providing color and depth descriptions of the scene at pixel and region level. The information given by these modules is fused in a mixture of experts fashion to improve the foreground detection accuracy. The main contributions of the paper are the region-based models of both background and foreground, built from the depth and color data. The obtained results using different database sequences demonstrate that the proposed approach leads to a higher detection accuracy with respect to existing state-of-the-art techniques.

Background foreground segmentation with RGB-D Kinect data: An efficient combination of classifiers

Low cost RGB-D cameras such as the Microsoft’s Kinect or the Asus’s Xtion Pro are completely changing the computer vision world, as they are being successfully used in several applications and research areas. Depth data are particularly attractive and suitable for applications based on moving objects detection through foreground/background segmentation approaches; the RGB-D applications proposed in literature employ, in general, state of the art foreground/background segmentation techniques based on the depth information without taking into account the color information. The novel approach that we propose is based on a combination of classifiers that allows improving background subtraction accuracy with respect to state of the art algorithms by jointly considering color and depth data. In particular, the combination of classifiers is based on a weighted average that allows to adaptively modifying the support of each classifier in the ensemble by considering foreground detections in the previous frames and the depth and color edges. In this way, it is possible to reduce false detections due to critical issues that can not be tackled by the individual classifiers such as: shadows and illumination changes, color and depth camouflage, moved background objects and noisy depth measurements. Moreover, we propose, for the best of the author’s knowledge, the first publicly available RGB-D benchmark dataset with hand-labeled ground truth of several challenging scenarios to test background/foreground segmentation algorithms.

Optimized HOG for on-road video based vehicle verification

Vision-based object detection from a moving platform becomes particularly challenging in the field of advanced driver assistance systems (ADAS). In this context, onboard vision-based vehicle verification strategies become critical, facing challenges derived from the variability of vehicles appearance, illumination, and vehicle speed. In this paper, an optimized HOG configuration for onboard vehicle verification is proposed which not only considers its spatial and orientation resolution, but descriptor processing strategies and classification. An in-depth analysis of the optimal settings for HOG for onboard vehicle verification is presented, in the context of SVM classification with different kernels. In contrast to many existing approaches, the evaluation is realized in a public and heterogeneous database of vehicle and non-vehicle images in different areas of the road, rendering excellent verification rates that outperform other similar approaches in the literature.

New definitions of margin for multi-class Boosting algorithms

La familia de algoritmos de Boosting son un tipo de técnicas de clasificación y regresión que han demostrado ser muy eficaces en problemas de Visión Computacional. Tal es el caso de los problemas de detección, de seguimiento o bien de reconocimiento de caras, personas, objetos deformables y acciones. El primer y más popular algoritmo de Boosting, AdaBoost, fue concebido para problemas binarios. Desde entonces, muchas han sido las propuestas que han aparecido con objeto de trasladarlo a otros dominios más generales: multiclase, multilabel, con costes, etc. Nuestro interés se centra en extender AdaBoost al terreno de la clasificación multiclase, considerándolo como un primer paso para posteriores ampliaciones. En la presente tesis proponemos dos algoritmos de Boosting para problemas multiclase basados en nuevas derivaciones del concepto margen. El primero de ellos, PIBoost, está concebido para abordar el problema descomponiéndolo en subproblemas binarios. Por un lado, usamos una codificación vectorial para representar etiquetas y, por otro, utilizamos la función de pérdida exponencial multiclase para evaluar las respuestas. Esta codificación produce un conjunto de valores margen que conllevan un rango de penalizaciones en caso de fallo y recompensas en caso de acierto. La optimización iterativa del modelo genera un proceso de Boosting asimétrico cuyos costes dependen del número de etiquetas separadas por cada clasificador débil. De este modo nuestro algoritmo de Boosting tiene en cuenta el desbalanceo debido a las clases a la hora de construir el clasificador. El resultado es un método bien fundamentado que extiende de manera canónica al AdaBoost original. El segundo algoritmo propuesto, BAdaCost, está concebido para problemas multiclase dotados de una matriz de costes. Motivados por los escasos trabajos dedicados a generalizar AdaBoost al terreno multiclase con costes, hemos propuesto un nuevo concepto de margen que, a su vez, permite derivar una función de pérdida adecuada para evaluar costes. Consideramos nuestro algoritmo como la extensión más canónica de AdaBoost para este tipo de problemas, ya que generaliza a los algoritmos SAMME, Cost-Sensitive AdaBoost y PIBoost. Por otro lado, sugerimos un simple procedimiento para calcular matrices de coste adecuadas para mejorar el rendimiento de Boosting a la hora de abordar problemas estándar y problemas con datos desbalanceados. Una serie de experimentos nos sirven para demostrar la efectividad de ambos métodos frente a otros conocidos algoritmos de Boosting multiclase en sus respectivas áreas. En dichos experimentos se usan bases de datos de referencia en el área de Machine Learning, en primer lugar para minimizar errores y en segundo lugar para minimizar costes. Además, hemos podido aplicar BAdaCost con éxito a un proceso de segmentación, un caso particular de problema con datos desbalanceados. Concluimos justificando el horizonte de futuro que encierra el marco de trabajo que presentamos, tanto por su aplicabilidad como por su flexibilidad teórica. Abstract The family of Boosting algorithms represents a type of classification and regression approach that has shown to be very effective in Computer Vision problems. Such is the case of detection, tracking and recognition of faces, people, deformable objects and actions. The first and most popular algorithm, AdaBoost, was introduced in the context of binary classification. Since then, many works have been proposed to extend it to the more general multi-class, multi-label, costsensitive, etc... domains. Our interest is centered in extending AdaBoost to two problems in the multi-class field, considering it a first step for upcoming generalizations. In this dissertation we propose two Boosting algorithms for multi-class classification based on new generalizations of the concept of margin. The first of them, PIBoost, is conceived to tackle the multi-class problem by solving many binary sub-problems. We use a vectorial codification to represent class labels and a multi-class exponential loss function to evaluate classifier responses. This representation produces a set of margin values that provide a range of penalties for failures and rewards for successes. The stagewise optimization of this model introduces an asymmetric Boosting procedure whose costs depend on the number of classes separated by each weak-learner. In this way the Boosting procedure takes into account class imbalances when building the ensemble. The resulting algorithm is a well grounded method that canonically extends the original AdaBoost. The second algorithm proposed, BAdaCost, is conceived for multi-class problems endowed with a cost matrix. Motivated by the few cost-sensitive extensions of AdaBoost to the multi-class field, we propose a new margin that, in turn, yields a new loss function appropriate for evaluating costs. Since BAdaCost generalizes SAMME, Cost-Sensitive AdaBoost and PIBoost algorithms, we consider our algorithm as a canonical extension of AdaBoost to this kind of problems. We additionally suggest a simple procedure to compute cost matrices that improve the performance of Boosting in standard and unbalanced problems. A set of experiments is carried out to demonstrate the effectiveness of both methods against other relevant Boosting algorithms in their respective areas. In the experiments we resort to benchmark data sets used in the Machine Learning community, firstly for minimizing classification errors and secondly for minimizing costs. In addition, we successfully applied BAdaCost to a segmentation task, a particular problem in presence of imbalanced data. We conclude the thesis justifying the horizon of future improvements encompassed in our framework, due to its applicability and theoretical flexibility.

Bayesian models for finding and grouping junctions

In this paper, we propose two Bayesian methods for detecting and grouping junctions. Our junction detection method evolves from the Kona approach, and it is based on a competitive greedy procedure inspired in the region competition method. Then, junction grouping is accomplished by finding connecting paths between pairs of junctions. Path searching is performed by applying a Bayesian A* algorithm that has been recently proposed. Both methods are efficient and robust, and they are tested with synthetic and real images.

Real-time 3D semi-local surface patch extraction using GPGPU

Feature vectors can be anything from simple surface normals to more complex feature descriptors. Feature extraction is important to solve various computer vision problems: e.g. registration, object recognition and scene understanding. Most of these techniques cannot be computed online due to their complexity and the context where they are applied. Therefore, computing these features in real-time for many points in the scene is impossible. In this work, a hardware-based implementation of 3D feature extraction and 3D object recognition is proposed to accelerate these methods and therefore the entire pipeline of RGBD based computer vision systems where such features are typically used. The use of a GPU as a general purpose processor can achieve considerable speed-ups compared with a CPU implementation. In this work, advantageous results are obtained using the GPU to accelerate the computation of a 3D descriptor based on the calculation of 3D semi-local surface patches of partial views. This allows descriptor computation at several points of a scene in real-time. Benefits of the accelerated descriptor have been demonstrated in object recognition tasks. Source code will be made publicly available as contribution to the Open Source Point Cloud Library.

Real time motion estimation using a neural architecture implemented on GPUs

This work describes a neural network based architecture that represents and estimates object motion in videos. This architecture addresses multiple computer vision tasks such as image segmentation, object representation or characterization, motion analysis and tracking. The use of a neural network architecture allows for the simultaneous estimation of global and local motion and the representation of deformable objects. This architecture also avoids the problem of finding corresponding features while tracking moving objects. Due to the parallel nature of neural networks, the architecture has been implemented on GPUs that allows the system to meet a set of requirements such as: time constraints management, robustness, high processing speed and re-configurability. Experiments are presented that demonstrate the validity of our architecture to solve problems of mobile agents tracking and motion analysis.

RGBD Human-Hand recognition for the Interaction with Robot-Hand

New low cost sensors and the new open free libraries for 3D image processing are permitting to achieve important advances for robot vision applications such as tridimensional object recognition, semantic mapping, navigation and localization of robots, human detection and/or gesture recognition for human-machine interaction. In this paper, a method to recognize the human hand and to track the fingers is proposed. This new method is based on point clouds from range images, RGBD. It does not require visual marks, camera calibration, environment knowledge and complex expensive acquisition systems. Furthermore, this method has been implemented to create a human interface in order to move a robot hand. The human hand is recognized and the movement of the fingers is analyzed. Afterwards, it is imitated from a Barret hand, using communication events programmed from ROS.

Three-dimensional planar model estimation using multi-constraint knowledge based on k-means and RANSAC

Plane model extraction from three-dimensional point clouds is a necessary step in many different applications such as planar object reconstruction, indoor mapping and indoor localization. Different RANdom SAmple Consensus (RANSAC)-based methods have been proposed for this purpose in recent years. In this study, we propose a novel method-based on RANSAC called Multiplane Model Estimation, which can estimate multiple plane models simultaneously from a noisy point cloud using the knowledge extracted from a scene (or an object) in order to reconstruct it accurately. This method comprises two steps: first, it clusters the data into planar faces that preserve some constraints defined by knowledge related to the object (e.g., the angles between faces); and second, the models of the planes are estimated based on these data using a novel multi-constraint RANSAC. We performed experiments in the clustering and RANSAC stages, which showed that the proposed method performed better than state-of-the-art methods.

A Novel Prediction Method for Early Recognition of Global Human Behaviour in Image Sequences

Human behaviour recognition has been, and still remains, a challenging problem that involves different areas of computational intelligence. The automated understanding of people activities from video sequences is an open research topic in which the computer vision and pattern recognition areas have made big efforts. In this paper, the problem is studied from a prediction point of view. We propose a novel method able to early detect behaviour using a small portion of the input, in addition to the capabilities of it to predict behaviour from new inputs. Specifically, we propose a predictive method based on a simple representation of trajectories of a person in the scene which allows a high level understanding of the global human behaviour. The representation of the trajectory is used as a descriptor of the activity of the individual. The descriptors are used as a cue of a classification stage for pattern recognition purposes. Classifiers are trained using the trajectory representation of the complete sequence. However, partial sequences are processed to evaluate the early prediction capabilities having a specific observation time of the scene. The experiments have been carried out using the three different dataset of the CAVIAR database taken into account the behaviour of an individual. Additionally, different classic classifiers have been used for experimentation in order to evaluate the robustness of the proposal. Results confirm the high accuracy of the proposal on the early recognition of people behaviours.

Sperimentazione con algoritmi per l'analisi di nuvole di punti per applicazioni di guida autonoma

Al giorno d’oggi quasi tutte le persone possiedono un mezzo motorizzato che utilizzano per spostarsi. Tale operazione, che risulta semplice per una persona, può essere compiuta da un robot o un autoveicolo in modo autonomo? La risposta a questa domanda è si, ma se ad una persona serve solo un po’ di pratica per guidare, questa azione non risulta altrettanto immediata per dei veicoli motorizzati. In soccorso ad essi vi è la Computer Vision, un ramo dell’informatica che, in un certo senso, rende un elaboratore elettronico in grado di percepire l’ambiente circostante, nel modo in cui una persona fa con i propri occhi. Oggi ci concentreremo su due campi della computer vision, lo SLAM o Simultaneous Localization and Mapping, che rende un robot in grado di mappare, attraverso una camera, il mondo in cui si trova ed allo stesso tempo di localizzare, istante per istante, la propria posizione all’interno di esso, e la Plane Detection, che permette di estrapolare i piani presenti all’interno di una data immagine.