A Taxonomy of image matching techniques for stereo vision

Stereo vision is a method of depth perception, in which depth information is inferred from two (or more) images of a scene, taken from different perspectives. Applications of stereo vision include aerial photogrammetry, autonomous vehicle guidance, robotics, industrial automation and stereomicroscopy. A key issue in stereo vision is that of image matching, or identifying corresponding points in a stereo pair. The difference in the positions of corresponding points in image coordinates is termed the parallax or disparity. When the orientation of the two cameras is known, corresponding points may be projected back to find the location of the original object point in world coordinates. Matching techniques are typically categorised according to the nature of the matching primitives they use and the matching strategy they employ. This report provides a detailed taxonomy of image matching techniques, including area based, transform based, feature based, phase based, hybrid, relaxation based, dynamic programming and object space methods. A number of area based matching metrics as well as the rank and census transforms were implemented, in order to investigate their suitability for a real-time stereo sensor for mining automation applications. The requirements of this sensor were speed, robustness, and the ability to produce a dense depth map. The Sum of Absolute Differences matching metric was the least computationally expensive; however, this metric was the most sensitive to radiometric distortion. Metrics such as the Zero Mean Sum of Absolute Differences and Normalised Cross Correlation were the most robust to this type of distortion but introduced additional computational complexity. The rank and census transforms were found to be robust to radiometric distortion, in addition to having low computational complexity. They are therefore prime candidates for a matching algorithm for a stereo sensor for real-time mining applications. A number of issues came to light during this investigation which may merit further work. These include devising a means to evaluate and compare disparity results of different matching algorithms, and finding a method of assigning a level of confidence to a match. Another issue of interest is the possibility of statistically combining the results of different matching algorithms, in order to improve robustness.

A mask-based approach for the geometric calibration of thermal-infrared cameras

Accurate and efficient thermal-infrared (IR) camera calibration is important for advancing computer vision research within the thermal modality. This paper presents an approach for geometrically calibrating individual and multiple cameras in both the thermal and visible modalities. The proposed technique can be used to correct for lens distortion and to simultaneously reference both visible and thermal-IR cameras to a single coordinate frame. The most popular existing approach for the geometric calibration of thermal cameras uses a printed chessboard heated by a flood lamp and is comparatively inaccurate and difficult to execute. Additionally, software toolkits provided for calibration either are unsuitable for this task or require substantial manual intervention. A new geometric mask with high thermal contrast and not requiring a flood lamp is presented as an alternative calibration pattern. Calibration points on the pattern are then accurately located using a clustering-based algorithm which utilizes the maximally stable extremal region detector. This algorithm is integrated into an automatic end-to-end system for calibrating single or multiple cameras. The evaluation shows that using the proposed mask achieves a mean reprojection error up to 78% lower than that using a heated chessboard. The effectiveness of the approach is further demonstrated by using it to calibrate two multiple-camera multiple-modality setups. Source code and binaries for the developed software are provided on the project Web site.

Deformable face ensemble alignment with robust grouped-L1 anchors

Many methods exist at the moment for deformable face fitting. A drawback to nearly all these approaches is that they are (i) noisy in terms of landmark positions, and (ii) the noise is biased across frames (i.e. the misalignment is toward common directions across all frames). In this paper we propose a grouped $\mathcal{L}1$-norm anchored method for simultaneously aligning an ensemble of deformable face images stemming from the same subject, given noisy heterogeneous landmark estimates. Impressive alignment performance improvement and refinement is obtained using very weak initialization as "anchors".

Nonrigid face alignment for unknown subject in video

Non-rigid face alignment is a very important task in a large range of applications but the existing tracking based non-rigid face alignment methods are either inaccurate or requiring person-specific model. This dissertation has developed simultaneous alignment algorithms that overcome these constraints and provide alignment with high accuracy, efficiency, robustness to varying image condition, and requirement of only generic model.

Matching image sets via adaptive multi convex hull

Traditional nearest points methods use all the samples in an image set to construct a single convex or affine hull model for classification. However, strong artificial features and noisy data may be generated from combinations of training samples when significant intra-class variations and/or noise occur in the image set. Existing multi-model approaches extract local models by clustering each image set individually only once, with fixed clusters used for matching with various image sets. This may not be optimal for discrimination, as undesirable environmental conditions (eg. illumination and pose variations) may result in the two closest clusters representing different characteristics of an object (eg. frontal face being compared to non-frontal face). To address the above problem, we propose a novel approach to enhance nearest points based methods by integrating affine/convex hull classification with an adapted multi-model approach. We first extract multiple local convex hulls from a query image set via maximum margin clustering to diminish the artificial variations and constrain the noise in local convex hulls. We then propose adaptive reference clustering (ARC) to constrain the clustering of each gallery image set by forcing the clusters to have resemblance to the clusters in the query image set. By applying ARC, noisy clusters in the query set can be discarded. Experiments on Honda, MoBo and ETH-80 datasets show that the proposed method outperforms single model approaches and other recent techniques, such as Sparse Approximated Nearest Points, Mutual Subspace Method and Manifold Discriminant Analysis.

Improved image set classification via joint sparse approximated nearest subspaces

Existing multi-model approaches for image set classification extract local models by clustering each image set individually only once, with fixed clusters used for matching with other image sets. However, this may result in the two closest clusters to represent different characteristics of an object, due to different undesirable environmental conditions (such as variations in illumination and pose). To address this problem, we propose to constrain the clustering of each query image set by forcing the clusters to have resemblance to the clusters in the gallery image sets. We first define a Frobenius norm distance between subspaces over Grassmann manifolds based on reconstruction error. We then extract local linear subspaces from a gallery image set via sparse representation. For each local linear subspace, we adaptively construct the corresponding closest subspace from the samples of a probe image set by joint sparse representation. We show that by minimising the sparse representation reconstruction error, we approach the nearest point on a Grassmann manifold. Experiments on Honda, ETH-80 and Cambridge-Gesture datasets show that the proposed method consistently outperforms several other recent techniques, such as Affine Hull based Image Set Distance (AHISD), Sparse Approximated Nearest Points (SANP) and Manifold Discriminant Analysis (MDA).

Automated sensory data alignment for environmental and epidermal change monitoring

In this paper we present research adapting a state of the art condition-invariant robotic place recognition algorithm to the role of automated inter- and intra-image alignment of sensor observations of environmental and skin change over time. The approach involves inverting the typical criteria placed upon navigation algorithms in robotics; we exploit rather than attempt to fix the limited camera viewpoint invariance of such algorithms, showing that approximate viewpoint repetition is realistic in a wide range of environments and medical applications. We demonstrate the algorithms automatically aligning challenging visual data from a range of real-world applications: ecological monitoring of environmental change, aerial observation of natural disasters including flooding, tsunamis and bushfires and tracking wound recovery and sun damage over time and present a prototype active guidance system for enforcing viewpoint repetition. We hope to provide an interesting case study for how traditional research criteria in robotics can be inverted to provide useful outcomes in applied situations.

Evaluation of object detection proposals under condition variations

Object detection is a fundamental task in many computer vision applications, therefore the importance of evaluating the quality of object detection is well acknowledged in this domain. This process gives insight into the capabilities of methods in handling environmental changes. In this paper, a new method for object detection is introduced that combines the Selective Search and EdgeBoxes. We tested these three methods under environmental variations. Our experiments demonstrate the outperformance of the combination method under illumination and view point variations.

Improving deep convultional neural networks with unsupervised feature learning

The latest generation of Deep Convolutional Neural Networks (DCNN) have dramatically advanced challenging computer vision tasks, especially in object detection and object classification, achieving state-of-the-art performance in several computer vision tasks including text recognition, sign recognition, face recognition and scene understanding. The depth of these supervised networks has enabled learning deeper and hierarchical representation of features. In parallel, unsupervised deep learning such as Convolutional Deep Belief Network (CDBN) has also achieved state-of-the-art in many computer vision tasks. However, there is very limited research on jointly exploiting the strength of these two approaches. In this paper, we investigate the learning capability of both methods. We compare the output of individual layers and show that many learnt filters and outputs of the corresponding level layer are almost similar for both approaches. Stacking the DCNN on top of unsupervised layers or replacing layers in the DCNN with the corresponding learnt layers in the CDBN can improve the recognition/classification accuracy and training computational expense. We demonstrate the validity of the proposal on ImageNet dataset.

Unsupervised Domain Adaptation by Domain Invariant Projection

Domain-invariant representations are key to addressing the domain shift problem where the training and test exam- ples follow different distributions. Existing techniques that have attempted to match the distributions of the source and target domains typically compare these distributions in the original feature space. This space, however, may not be di- rectly suitable for such a comparison, since some of the fea- tures may have been distorted by the domain shift, or may be domain specific. In this paper, we introduce a Domain Invariant Projection approach: An unsupervised domain adaptation method that overcomes this issue by extracting the information that is invariant across the source and tar- get domains. More specifically, we learn a projection of the data to a low-dimensional latent space where the distance between the empirical distributions of the source and target examples is minimized. We demonstrate the effectiveness of our approach on the task of visual object recognition and show that it outperforms state-of-the-art methods on a stan- dard domain adaptation benchmark dataset

Automatically Recovering Geometry and Texture from Large Sets of Calibrated Images

Three-dimensional models which contain both geometry and texture have numerous applications such as urban planning, physical simulation, and virtual environments. A major focus of computer vision (and recently graphics) research is the automatic recovery of three-dimensional models from two-dimensional images. After many years of research this goal is yet to be achieved. Most practical modeling systems require substantial human input and unlike automatic systems are not scalable. This thesis presents a novel method for automatically recovering dense surface patches using large sets (1000's) of calibrated images taken from arbitrary positions within the scene. Physical instruments, such as Global Positioning System (GPS), inertial sensors, and inclinometers, are used to estimate the position and orientation of each image. Essentially, the problem is to find corresponding points in each of the images. Once a correspondence has been established, calculating its three-dimensional position is simply a matter of geometry. Long baseline images improve the accuracy. Short baseline images and the large number of images greatly simplifies the correspondence problem. The initial stage of the algorithm is completely local and scales linearly with the number of images. Subsequent stages are global in nature, exploit geometric constraints, and scale quadratically with the complexity of the underlying scene. We describe techniques for: 1) detecting and localizing surface patches; 2) refining camera calibration estimates and rejecting false positive surfels; and 3) grouping surface patches into surfaces and growing the surface along a two-dimensional manifold. We also discuss a method for producing high quality, textured three-dimensional models from these surfaces. Some of the most important characteristics of this approach are that it: 1) uses and refines noisy calibration estimates; 2) compensates for large variations in illumination; 3) tolerates significant soft occlusion (e.g. tree branches); and 4) associates, at a fundamental level, an estimated normal (i.e. no frontal-planar assumption) and texture with each surface patch.

Robust and Efficient 3D Recognition by Alignment

Alignment is a prevalent approach for recognizing 3D objects in 2D images. A major problem with current implementations is how to robustly handle errors that propagate from uncertainties in the locations of image features. This thesis gives a technique for bounding these errors. The technique makes use of a new solution to the problem of recovering 3D pose from three matching point pairs under weak-perspective projection. Furthermore, the error bounds are used to demonstrate that using line segments for features instead of points significantly reduces the false positive rate, to the extent that alignment can remain reliable even in cluttered scenes.

A Modern Differential Geometric Approach to Shape from Shading

How the visual system extracts shape information from a single grey-level image can be approached by examining how the information about shape is contained in the image. This technical report considers the characteristic equations derived by Horn as a dynamical system. Certain image critical points generate dynamical system critical points. The stable and unstable manifolds of these critical points correspond to convex and concave solution surfaces, giving more general existence and uniqueness results. A new kind of highly parallel, robust shape from shading algorithm is suggested on neighborhoods of these critical points. The information at bounding contours in the image is also analyzed.