Automated object identification and position estimation for airport lighting quality assessment

The development of an automated system for the quality assessment of aerodrome ground lighting (AGL), in accordance with associated standards and recommendations, is presented. The system is composed of an image sensor, placed inside the cockpit of an aircraft to record images of the AGL during a normal descent to an aerodrome. A model-based methodology is used to ascertain the optimum match between a template of the AGL and the actual image data in order to calculate the position and orientation of the camera at the instant the image was acquired. The camera position and orientation data are used along with the pixel grey level for each imaged luminaire, to estimate a value for the luminous intensity of a given luminaire. This can then be compared with the expected brightness for that luminaire to ensure it is operating to the required standards. As such, a metric for the quality of the AGL pattern is determined. Experiments on real image data is presented to demonstrate the application and effectiveness of the system.

Modelling 3D camera movement for vibration characterisation and multiple object identification with application to lighting assessment

Utilising cameras as a means to survey the surrounding environment is becoming increasingly popular in a number of different research areas and applications. Central to using camera sensors as input to a vision system, is the need to be able to manipulate and process the information captured in these images. One such application, is the use of cameras to monitor the quality of airport landing lighting at aerodromes where a camera is placed inside an aircraft and used to record images of the lighting pattern during the landing phase of a flight. The images are processed to determine a performance metric. This requires the development of custom software for the localisation and identification of luminaires within the image data. However, because of the necessity to keep airport operations functioning as efficiently as possible, it is difficult to collect enough image data to develop, test and validate any developed software. In this paper, we present a technique to model a virtual landing lighting pattern. A mathematical model is postulated which represents the glide path of the aircraft including random deviations from the expected path. A morphological method has been developed to localise and track the luminaires under different operating conditions. © 2011 IEEE.

Role of early visual cortex in trans-saccadic memory of object features

Early visual cortex (EVC) participates in visual feature memory and the updating of remembered locations across saccades, but its role in the trans-saccadic integration of object features is unknown. We hypothesized that if EVC is involved in updating object features relative to gaze, feature memory should be disrupted when saccades remap an object representation into a simultaneously perturbed EVC site. To test this, we applied transcranial magnetic stimulation (TMS) over functional magnetic resonance imaging-localized EVC clusters corresponding to the bottom left/right visual quadrants (VQs). During experiments, these VQs were probed psychophysically by briefly presenting a central object (Gabor patch) while subjects fixated gaze to the right or left (and above). After a short memory interval, participants were required to detect the relative change in orientation of a re-presented test object at the same spatial location. Participants either sustained fixation during the memory interval (fixation task) or made a horizontal saccade that either maintained or reversed the VQ of the object (saccade task). Three TMS pulses (coinciding with the pre-, peri-, and postsaccade intervals) were applied to the left or right EVC. This had no effect when (a) fixation was maintained, (b) saccades kept the object in the same VQ, or (c) the EVC quadrant corresponding to the first object was stimulated. However, as predicted, TMS reduced performance when saccades (especially larger saccades) crossed the remembered object location and brought it into the VQ corresponding to the TMS site. This suppression effect was statistically significant for leftward saccades and followed a weaker trend for rightward saccades. These causal results are consistent with the idea that EVC is involved in the gaze-centered updating of object features for trans-saccadic memory and perception.

Cortical object segregation and categorization by multi-scale line and edge coding

In this paper we present an improved scheme for line and edge detection in cortical area V1, based on responses of simple and complex cells, truly multi-scale with no free parameters. We illustrate the multi-scale representation for visual reconstruction, and show how object segregation can be achieved with coarse-to-finescale groupings. A two-level object categorization scenario is tested in which pre-categorization is based on coarse scales only, and final categorization on coarse plus fine scales. Processing schemes are discussed in the framework of a complete cortical architecture.

Multi-scale keypoint hierarchy for Focus-of-Attention and object detection

Hypercolumns in area V1 contain frequency- and orientation-selective simple and complex cells for line (bar) and edge coding, plus end-stopped cells for key- point (vertex) detection. A single-scale (single-frequency) mathematical model of single and double end-stopped cells on the basis of Gabor filter responses was developed by Heitger et al. (1992 Vision Research 32 963-981). We developed an improved model by stabilising keypoint detection over neighbouring micro- scales.

Invariant multi-scale object categorisation and recognition

Object recognition requires that templates with canonical views are stored in memory. Such templates must somehow be normalised. In this paper we present a novel method for obtaining 2D translation, rotation and size invariance. Cortical simple, complex and end-stopped cells provide multi-scale maps of lines, edges and keypoints. These maps are combined such that objects are characterised. Dynamic routing in neighbouring neural layers allows feature maps of input objects and stored templates to converge. We illustrate the construction of group templates and the invariance method for object categorisation and recognition in the context of a cortical architecture, which can be applied in computer vision.

Integrated multi-scale architecture of the cortex with application to computer vision

Tese de dout., Engenharia Electrónica e de Computadores, Faculdade de Ciência e Tecnologia, Universidade do Algarve, 2007

A vision system for detecting paths and moving obstacles for the blind

In this paper we present a monocular vision system for a navigation aid. The system assists blind persons in following paths and sidewalks, and it alerts the user to moving obstacles which may be on collision course. Path borders and the vanishing point are de-tected by edges and an adapted Hough transform. Opti-cal flow is detected by using a hierarchical, multi-scale tree structure with annotated keypoints. The tree struc-ture also allows to segregate moving objects, indicating where on the path the objects are. Moreover, the centre of the object relative to the vanishing point indicates whether an object is approaching or not.

Biologically Inspired Vision for Indoor Robot Navigation

Ultrasonic, infrared, laser and other sensors are being applied in robotics. Although combinations of these have allowed robots to navigate, they are only suited for specific scenarios, depending on their limitations. Recent advances in computer vision are turning cameras into useful low-cost sensors that can operate in most types of environments. Cameras enable robots to detect obstacles, recognize objects, obtain visual odometry, detect and recognize people and gestures, among other possibilities. In this paper we present a completely biologically inspired vision system for robot navigation. It comprises stereo vision for obstacle detection, and object recognition for landmark-based navigation. We employ a novel keypoint descriptor which codes responses of cortical complex cells. We also present a biologically inspired saliency component, based on disparity and colour.

Real-Time Automatic Object Classification and Tracking using Genetic Programming and NVIDIA R CUDA TM

Genetic Programming (GP) is a widely used methodology for solving various computational problems. GP's problem solving ability is usually hindered by its long execution times. In this thesis, GP is applied toward real-time computer vision. In particular, object classification and tracking using a parallel GP system is discussed. First, a study of suitable GP languages for object classification is presented. Two main GP approaches for visual pattern classification, namely the block-classifiers and the pixel-classifiers, were studied. Results showed that the pixel-classifiers generally performed better. Using these results, a suitable language was selected for the real-time implementation. Synthetic video data was used in the experiments. The goal of the experiments was to evolve a unique classifier for each texture pattern that existed in the video. The experiments revealed that the system was capable of correctly tracking the textures in the video. The performance of the system was on-par with real-time requirements.

Training deep convolutional architectures for vision

Les tâches de vision artificielle telles que la reconnaissance d’objets demeurent irrésolues à ce jour. Les algorithmes d’apprentissage tels que les Réseaux de Neurones Artificiels (RNA), représentent une approche prometteuse permettant d’apprendre des caractéristiques utiles pour ces tâches. Ce processus d’optimisation est néanmoins difficile. Les réseaux profonds à base de Machine de Boltzmann Restreintes (RBM) ont récemment été proposés afin de guider l’extraction de représentations intermédiaires, grâce à un algorithme d’apprentissage non-supervisé. Ce mémoire présente, par l’entremise de trois articles, des contributions à ce domaine de recherche. Le premier article traite de la RBM convolutionelle. L’usage de champs réceptifs locaux ainsi que le regroupement d’unités cachées en couches partageant les même paramètres, réduit considérablement le nombre de paramètres à apprendre et engendre des détecteurs de caractéristiques locaux et équivariant aux translations. Ceci mène à des modèles ayant une meilleure vraisemblance, comparativement aux RBMs entraînées sur des segments d’images. Le deuxième article est motivé par des découvertes récentes en neurosciences. Il analyse l’impact d’unités quadratiques sur des tâches de classification visuelles, ainsi que celui d’une nouvelle fonction d’activation. Nous observons que les RNAs à base d’unités quadratiques utilisant la fonction softsign, donnent de meilleures performances de généralisation. Le dernière article quand à lui, offre une vision critique des algorithmes populaires d’entraînement de RBMs. Nous montrons que l’algorithme de Divergence Contrastive (CD) et la CD Persistente ne sont pas robustes : tous deux nécessitent une surface d’énergie relativement plate afin que leur chaîne négative puisse mixer. La PCD à "poids rapides" contourne ce problème en perturbant légèrement le modèle, cependant, ceci génère des échantillons bruités. L’usage de chaînes tempérées dans la phase négative est une façon robuste d’adresser ces problèmes et mène à de meilleurs modèles génératifs.

Exploitation de contraintes photométriques et géométriques en vision : application au suivi, au calibrage et à la reconstruction

Cette thése a été réalisée dans le cadre d'une cotutelle avec l'Institut National Polytechnique de Grenoble (France). La recherche a été effectuée au sein des laboratoires de vision 3D (DIRO, UdM) et PERCEPTION-INRIA (Grenoble).

Deep learning of representations and its application to computer vision

L’objectif de cette thèse par articles est de présenter modestement quelques étapes du parcours qui mènera (on espère) à une solution générale du problème de l’intelligence artificielle. Cette thèse contient quatre articles qui présentent chacun une différente nouvelle méthode d’inférence perceptive en utilisant l’apprentissage machine et, plus particulièrement, les réseaux neuronaux profonds. Chacun de ces documents met en évidence l’utilité de sa méthode proposée dans le cadre d’une tâche de vision par ordinateur. Ces méthodes sont applicables dans un contexte plus général, et dans certains cas elles on tété appliquées ailleurs, mais ceci ne sera pas abordé dans le contexte de cette de thèse. Dans le premier article, nous présentons deux nouveaux algorithmes d’inférence variationelle pour le modèle génératif d’images appelé codage parcimonieux “spike- and-slab” (CPSS). Ces méthodes d’inférence plus rapides nous permettent d’utiliser des modèles CPSS de tailles beaucoup plus grandes qu’auparavant. Nous démontrons qu’elles sont meilleures pour extraire des détecteur de caractéristiques quand très peu d’exemples étiquetés sont disponibles pour l’entraînement. Partant d’un modèle CPSS, nous construisons ensuite une architecture profonde, la machine de Boltzmann profonde partiellement dirigée (MBP-PD). Ce modèle a été conçu de manière à simplifier d’entraînement des machines de Boltzmann profondes qui nécessitent normalement une phase de pré-entraînement glouton pour chaque couche. Ce problème est réglé dans une certaine mesure, mais le coût d’inférence dans le nouveau modèle est relativement trop élevé pour permettre de l’utiliser de manière pratique. Dans le deuxième article, nous revenons au problème d’entraînement joint de machines de Boltzmann profondes. Cette fois, au lieu de changer de famille de modèles, nous introduisons un nouveau critère d’entraînement qui donne naissance aux machines de Boltzmann profondes à multiples prédictions (MBP-MP). Les MBP-MP sont entraînables en une seule étape et ont un meilleur taux de succès en classification que les MBP classiques. Elles s’entraînent aussi avec des méthodes variationelles standard au lieu de nécessiter un classificateur discriminant pour obtenir un bon taux de succès en classification. Par contre, un des inconvénients de tels modèles est leur incapacité de générer deséchantillons, mais ceci n’est pas trop grave puisque la performance de classification des machines de Boltzmann profondes n’est plus une priorité étant donné les dernières avancées en apprentissage supervisé. Malgré cela, les MBP-MP demeurent intéressantes parce qu’elles sont capable d’accomplir certaines tâches que des modèles purement supervisés ne peuvent pas faire, telles que celle de classifier des données incomplètes ou encore celle de combler intelligemment l’information manquante dans ces données incomplètes. Le travail présenté dans cette thèse s’est déroulé au milieu d’une période de transformations importantes du domaine de l’apprentissage à réseaux neuronaux profonds qui a été déclenchée par la découverte de l’algorithme de “dropout” par Geoffrey Hinton. Dropout rend possible un entraînement purement supervisé d’architectures de propagation unidirectionnel sans être exposé au danger de sur- entraînement. Le troisième article présenté dans cette thèse introduit une nouvelle fonction d’activation spécialement con ̧cue pour aller avec l’algorithme de Dropout. Cette fonction d’activation, appelée maxout, permet l’utilisation de aggrégation multi-canal dans un contexte d’apprentissage purement supervisé. Nous démontrons comment plusieurs tâches de reconnaissance d’objets sont mieux accomplies par l’utilisation de maxout. Pour terminer, sont présentons un vrai cas d’utilisation dans l’industrie pour la transcription d’adresses de maisons à plusieurs chiffres. En combinant maxout avec une nouvelle sorte de couche de sortie pour des réseaux neuronaux de convolution, nous démontrons qu’il est possible d’atteindre un taux de succès comparable à celui des humains sur un ensemble de données coriace constitué de photos prises par les voitures de Google. Ce système a été déployé avec succès chez Google pour lire environ cent million d’adresses de maisons.

Hierarchical Object Recognition Using Libraries of Parameterized Model Sub-Parts

This thesis describes the development of a model-based vision system that exploits hierarchies of both object structure and object scale. The focus of the research is to use these hierarchies to achieve robust recognition based on effective organization and indexing schemes for model libraries. The goal of the system is to recognize parameterized instances of non-rigid model objects contained in a large knowledge base despite the presence of noise and occlusion. Robustness is achieved by developing a system that can recognize viewed objects that are scaled or mirror-image instances of the known models or that contain components sub-parts with different relative scaling, rotation, or translation than in models. The approach taken in this thesis is to develop an object shape representation that incorporates a component sub-part hierarchy- to allow for efficient and correct indexing into an automatically generated model library as well as for relative parameterization among sub-parts, and a scale hierarchy- to allow for a general to specific recognition procedure. After analysis of the issues and inherent tradeoffs in the recognition process, a system is implemented using a representation based on significant contour curvature changes and a recognition engine based on geometric constraints of feature properties. Examples of the system's performance are given, followed by an analysis of the results. In conclusion, the system's benefits and limitations are presented.

View-Based Strategies for 3D Object Recognition

A persistent issue of debate in the area of 3D object recognition concerns the nature of the experientially acquired object models in the primate visual system. One prominent proposal in this regard has expounded the use of object centered models, such as representations of the objects' 3D structures in a coordinate frame independent of the viewing parameters [Marr and Nishihara, 1978]. In contrast to this is another proposal which suggests that the viewing parameters encountered during the learning phase might be inextricably linked to subsequent performance on a recognition task [Tarr and Pinker, 1989; Poggio and Edelman, 1990]. The 'object model', according to this idea, is simply a collection of the sample views encountered during training. Given that object centered recognition strategies have the attractive feature of leading to viewpoint independence, they have garnered much of the research effort in the field of computational vision. Furthermore, since human recognition performance seems remarkably robust in the face of imaging variations [Ellis et al., 1989], it has often been implicitly assumed that the visual system employs an object centered strategy. In the present study we examine this assumption more closely. Our experimental results with a class of novel 3D structures strongly suggest the use of a view-based strategy by the human visual system even when it has the opportunity of constructing and using object-centered models. In fact, for our chosen class of objects, the results seem to support a stronger claim: 3D object recognition is 2D view-based.