Learning and Example Selection for Object and Pattern Detection

This thesis presents a learning based approach for detecting classes of objects and patterns with variable image appearance but highly predictable image boundaries. It consists of two parts. In part one, we introduce our object and pattern detection approach using a concrete human face detection example. The approach first builds a distribution-based model of the target pattern class in an appropriate feature space to describe the target's variable image appearance. It then learns from examples a similarity measure for matching new patterns against the distribution-based target model. The approach makes few assumptions about the target pattern class and should therefore be fairly general, as long as the target class has predictable image boundaries. Because our object and pattern detection approach is very much learning-based, how well a system eventually performs depends heavily on the quality of training examples it receives. The second part of this thesis looks at how one can select high quality examples for function approximation learning tasks. We propose an {em active learning} formulation for function approximation, and show for three specific approximation function classes, that the active example selection strategy learns its target with fewer data samples than random sampling. We then simplify the original active learning formulation, and show how it leads to a tractable example selection paradigm, suitable for use in many object and pattern detection problems.

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.

Exploring visual features through Gabor representations for facial expression detection

Gabor representations have been widely used in facial analysis (face recognition, face detection and facial expression detection) due to their biological relevance and computational properties. Two popular Gabor representations used in literature are: 1) Log-Gabor and 2) Gabor energy filters. Even though these representations are somewhat similar, they also have distinct differences as the Log-Gabor filters mimic the simple cells in the visual cortex while the Gabor energy filters emulate the complex cells, which causes subtle differences in the responses. In this paper, we analyze the difference between these two Gabor representations and quantify these differences on the task of facial action unit (AU) detection. In our experiments conducted on the Cohn-Kanade dataset, we report an average area underneath the ROC curve (A`) of 92.60% across 17 AUs for the Gabor energy filters, while the Log-Gabor representation achieved an average A` of 96.11%. This result suggests that small spatial differences that the Log-Gabor filters pick up on are more useful for AU detection than the differences in contours and edges that the Gabor energy filters extract.

Person-independent facial expression detection using constrained local models

In automatic facial expression detection, very accurate registration is desired which can be achieved via a deformable model approach where a dense mesh of 60-70 points on the face is used, such as an active appearance model (AAM). However, for applications where manually labeling frames is prohibitive, AAMs do not work well as they do not generalize well to unseen subjects. As such, a more coarse approach is taken for person-independent facial expression detection, where just a couple of key features (such as face and eyes) are tracked using a Viola-Jones type approach. The tracked image is normally post-processed to encode for shift and illumination invariance using a linear bank of filters. Recently, it was shown that this preprocessing step is of no benefit when close to ideal registration has been obtained. In this paper, we present a system based on the Constrained Local Model (CLM) which is a generic or person-independent face alignment algorithm which gains high accuracy. We show these results against the LBP feature extraction on the CK+ and GEMEP datasets.

Feature-domain super-resolution framework for gabor-based face and iris recognition

The low resolution of images has been one of the major limitations in recognising humans from a distance using their biometric traits, such as face and iris. Superresolution has been employed to improve the resolution and the recognition performance simultaneously, however the majority of techniques employed operate in the pixel domain, such that the biometric feature vectors are extracted from a super-resolved input image. Feature-domain superresolution has been proposed for face and iris, and is shown to further improve recognition performance by capitalising on direct super-resolving the features which are used for recognition. However, current feature-domain superresolution approaches are limited to simple linear features such as Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA), which are not the most discriminant features for biometrics. Gabor-based features have been shown to be one of the most discriminant features for biometrics including face and iris. This paper proposes a framework to conduct super-resolution in the non-linear Gabor feature domain to further improve the recognition performance of biometric systems. Experiments have confirmed the validity of the proposed approach, demonstrating superior performance to existing linear approaches for both face and iris biometrics.

Real-time video event detection in crowded scenes using MPEG derived features : a multiple instance learning approach

This paper presents an investigation into event detection in crowded scenes, where the event of interest co-occurs with other activities and only binary labels at the clip level are available. The proposed approach incorporates a fast feature descriptor from the MPEG domain, and a novel multiple instance learning (MIL) algorithm using sparse approximation and random sensing. MPEG motion vectors are used to build particle trajectories that represent the motion of objects in uniform video clips, and the MPEG DCT coefficients are used to compute a foreground map to remove background particles. Trajectories are transformed into the Fourier domain, and the Fourier representations are quantized into visual words using the K-Means algorithm. The proposed MIL algorithm models the scene as a linear combination of independent events, where each event is a distribution of visual words. Experimental results show that the proposed approaches achieve promising results for event detection compared to the state-of-the-art.

An efficient and robust system for multi-person event detection in real world indoor surveillance scenes

Due to the popularity of security cameras in public places, it is of interest to design an intelligent system that can efficiently detect events automatically. This paper proposes a novel algorithm for multi-person event detection. To ensure greater than real-time performance, features are extracted directly from compressed MPEG video. A novel histogram-based feature descriptor that captures the angles between extracted particle trajectories is proposed, which allows us to capture motion patterns of multi-person events in the video. To alleviate the need for fine-grained annotation, we propose the use of Labelled Latent Dirichlet Allocation, a “weakly supervised” method that allows the use of coarse temporal annotations which are much simpler to obtain. This novel system is able to run at approximately ten times real-time, while preserving state-of-theart detection performance for multi-person events on a 100-hour real-world surveillance dataset (TRECVid SED).

Autonomous UAVs wildlife detection using thermal imaging, predictive navigation and computer vision

There is an increased interest on the use of Unmanned Aerial Vehicles (UAVs) for wildlife and feral animal monitoring around the world. This paper describes a novel system which uses a predictive dynamic application that places the UAV ahead of a user, with a low cost thermal camera, a small onboard computer that identifies heat signatures of a target animal from a predetermined altitude and transmits that target’s GPS coordinates. A map is generated and various data sets and graphs are displayed using a GUI designed for easy use. The paper describes the hardware and software architecture and the probabilistic model for downward facing camera for the detection of an animal. Behavioral dynamics of target movement for the design of a Kalman filter and Markov model based prediction algorithm are used to place the UAV ahead of the user. Geometrical concepts and Haversine formula are applied to the maximum likelihood case in order to make a prediction regarding a future state of the user, thus delivering a new way point for autonomous navigation. Results show that the system is capable of autonomously locating animals from a predetermined height and generate a map showing the location of the animals ahead of the user.

Open source computer-vision based guidance system for UAVs on-board decision making

The use of UAVs for remote sensing tasks; e.g. agriculture, search and rescue is increasing. The ability for UAVs to autonomously find a target and perform on-board decision making, such as descending to a new altitude or landing next to a target is a desired capability. Computer-vision functionality allows the Unmanned Aerial Vehicle (UAV) to follow a designated flight plan, detect an object of interest, and change its planned path. In this paper we describe a low cost and an open source system where all image processing is achieved on-board the UAV using a Raspberry Pi 2 microprocessor interfaced with a camera. The Raspberry Pi and the autopilot are physically connected through serial and communicate via MAVProxy. The Raspberry Pi continuously monitors the flight path in real time through USB camera module. The algorithm checks whether the target is captured or not. If the target is detected, the position of the object in frame is represented in Cartesian coordinates and converted into estimate GPS coordinates. In parallel, the autopilot receives the target location approximate GPS and makes a decision to guide the UAV to a new location. This system also has potential uses in the field of Precision Agriculture, plant pest detection and disease outbreaks which cause detrimental financial damage to crop yields if not detected early on. Results show the algorithm is accurate to detect 99% of object of interest and the UAV is capable of navigation and doing on-board decision making.