Image segmentation and pose estimation of humans in video

This thesis introduces improved techniques towards automatically estimating the pose of humans from video. It examines a complete workflow to estimating pose, from the segmentation of the raw video stream to extract silhouettes, to using the silhouettes in order to determine the relative orientation of parts of the human body. The proposed segmentation algorithms have improved performance and reduced complexity, while the pose estimation shows superior accuracy during difficult cases of self occlusion.

Towards Autonomous Motion Vision

Earlier, we introduced a direct method called fixation for the recovery of shape and motion in the general case. The method uses neither feature correspondence nor optical flow. Instead, it directly employs the spatiotemporal gradients of image brightness. This work reports the experimental results of applying some of our fixation algorithms to a sequence of real images where the motion is a combination of translation and rotation. These results show that parameters such as the fization patch size have crucial effects on the estimation of some motion parameters. Some of the critical issues involved in the implementaion of our autonomous motion vision system are also discussed here. Among those are the criteria for automatic choice of an optimum size for the fixation patch, and an appropriate location for the fixation point which result in good estimates for important motion parameters. Finally, a calibration method is described for identifying the real location of the rotation axis in imaging systems.

Visual Tracking

A typical robot vision scenario might involve a vehicle moving with an unknown 3D motion (translation and rotation) while taking intensity images of an arbitrary environment. This paper describes the theory and implementation issues of tracking any desired point in the environment. This method is performed completely in software without any need to mechanically move the camera relative to the vehicle. This tracking technique is simple an inexpensive. Furthermore, it does not use either optical flow or feature correspondence. Instead, the spatio-temporal gradients of the input intensity images are used directly. The experimental results presented support the idea of tracking in software. The final result is a sequence of tracked images where the desired point is kept stationary in the images independent of the nature of the relative motion. Finally, the quality of these tracked images are examined using spatio-temporal gradient maps.

Robot Motion Vision by Fixation

In many motion-vision scenarios, a camera (mounted on a moving vehicle) takes images of an environment to find the "motion'' and shape. We introduce a direct-method called fixation for solving this motion-vision problem in its general case. Fixation uses neither feature-correspondence nor optical-flow. Instead, spatio-temporal brightness gradients are used directly. In contrast to previous direct methods, fixation does not restrict the motion or the environment. Moreover, fixation method neither requires tracked images as its input nor uses mechanical tracking for obtaining fixated images. The experimental results on real images are presented and the implementation issues and techniques are discussed.

A recursive approach to space resection using straight lines

An approach using straight lines as features to solve the photogrammetric space resection problem is presented. An explicit mathematical model relating straight lines, in both object and image space, is used. Based on this model, Kalman Filtering is applied to solve the space resection problem. The recursive property of the filter is used in an iterative process which uses the sequentially estimated camera location parameters to feedback to the feature extraction process in the image. This feedback process leads to a gradual reduction of the image space for feature searching, and consequently eliminates the bottleneck due to the high computational cost of the image segmentation phase. It also enables feature extraction and the determination of feature correspondence in image and object space in an automatic way, i.e., without operator interference. Results obtained from simulated and real data show that highly accurate space resection parameters are obtained as well as a progressive processing time reduction. The obtained accuracy, the automatic correspondence process, and the short related processing time show that the proposed approach can be used in many real-time machine vision systems, making possible the implementation of applications not feasible until now.

Pre-eruptive uplift in the Emeishan? [Correspondence]

In their correspondence, He and colleagues question our conclusion of little or no uplift preceding Emeishan volcanism that we reported in our letter1. Debate concerns the nature of the contact between the Maokou limestone and Emeishan volcanics, the depositional environment and volumetric significance of mafic hydromagmatic deposits (MHDs), and evidence for symmetrical domal thinning. MHDs in the Daqiao section are separated from the Maokou limestone by 100 m of subaerial basaltic lavas, but elsewhere MHDs — previously interpreted as basal conglomerates2, 3 — directly overlie the Maokou2, 3. MHDs thus feature strongly in basal sections of the Emeishan lava succession, as also recently shown4 elsewhere in the Emeishan. An irregular surface at the top of the Maokou limestone has been interpreted as an erosional unconformity2, 3, but clastic deposits presented as evidence of this erosion2, 3 are MHDs produced by explosive magma–water interaction1. A clear demonstration that this irregular top surface is an erosional truncation of limestone reef facies (slope/rim, flat, lagoonal) is currently lacking, but is critical because reefs and carbonate platforms show considerable natural relief of tens of metres. The persistent hot, wet climate since the Oligocene has produced well-developed weathering profiles on exposed Palaeozoic marine sedimentary sequences5, but weathering and karst relief of the uppermost Maokou limestone underlying the flood basalts have not been properly documented, nor shown to be of middle Permian age and immediately preceding emplacement of the large igneous province.

Wide baseline correspondence extraction beyond local features

Robust, affine covariant, feature extractors provide a means to extract correspondences between images captured by widely separated cameras. Advances in wide baseline correspondence extraction require looking beyond the robust feature extraction and matching approach. This study examines new techniques of extracting correspondences that take advantage of information contained in affine feature matches. Methods of improving the accuracy of a set of putative matches, eliminating incorrect matches and extracting large numbers of additional correspondences are explored. It is assumed that knowledge of the camera geometry is not available and not immediately recoverable. The new techniques are evaluated by means of an epipolar geometry estimation task. It is shown that these methods enable the computation of camera geometry in many cases where existing feature extractors cannot produce sufficient numbers of accurate correspondences.

Robust facial feature extraction and matching

Automated feature extraction and correspondence determination is an extremely important problem in the face recognition community as it often forms the foundation of the normalisation and database construction phases of many recognition and verification systems. This paper presents a completely automatic feature extraction system based upon a modified volume descriptor. These features form a stable descriptor for faces and are utilised in a reversible jump Markov chain Monte Carlo correspondence algorithm to automatically determine correspondences which exist between faces. The developed system is invariant to changes in pose and occlusion and results indicate that it is also robust to minor face deformations which may be present with variations in expression.

Synchronization of user-generated videos through trajectory correspondence and a refinement procedure

Temporal synchronization of multiple video recordings of the same dynamic event is a critical task in many computer vision applications e.g. novel view synthesis and 3D reconstruction. Typically this information is implied through the time-stamp information embedded in the video streams. User-generated videos shot using consumer grade equipment do not contain this information; hence, there is a need to temporally synchronize signals using the visual information itself. Previous work in this area has either assumed good quality data with relatively simple dynamic content or the availability of precise camera geometry. Our first contribution is a synchronization technique which tries to establish correspondence between feature trajectories across views in a novel way, and specifically targets the kind of complex content found in consumer generated sports recordings, without assuming precise knowledge of fundamental matrices or homographies. We evaluate performance using a number of real video recordings and show that our method is able to synchronize to within 1 sec, which is significantly better than previous approaches. Our second contribution is a robust and unsupervised view-invariant activity recognition descriptor that exploits recurrence plot theory on spatial tiles. The descriptor is individually shown to better characterize the activities from different views under occlusions than state-of-the-art approaches. We combine this descriptor with our proposed synchronization method and show that it can further refine the synchronization index. © 2013 ACM.

Modal Matching for Correspondence and Recognition

Modal matching is a new method for establishing correspondences and computing canonical descriptions. The method is based on the idea of describing objects in terms of generalized symmetries, as defined by each object's eigenmodes. The resulting modal description is used for object recognition and categorization, where shape similarities are expressed as the amounts of modal deformation energy needed to align the two objects. In general, modes provide a global-to-local ordering of shape deformation and thus allow for selecting which types of deformations are used in object alignment and comparison. In contrast to previous techniques, which required correspondence to be computed with an initial or prototype shape, modal matching utilizes a new type of finite element formulation that allows for an object's eigenmodes to be computed directly from available image information. This improved formulation provides greater generality and accuracy, and is applicable to data of any dimensionality. Correspondence results with 2-D contour and point feature data are shown, and recognition experiments with 2-D images of hand tools and airplanes are described.

A comparison of recent reanalysis datasets using objective feature tracking: Storm tracks and tropical easterly waves

Data from four recent reanalysis projects [ECMWF, NCEP-NCAR, NCEP - Department of Energy ( DOE), NASA] have been diagnosed at the scale of synoptic weather systems using an objective feature tracking method. The tracking statistics indicate that, overall, the reanalyses correspond very well in the Northern Hemisphere (NH) lower troposphere, although differences for the spatial distribution of mean intensities show that the ECMWF reanalysis is systematically stronger in the main storm track regions but weaker around major orographic features. A direct comparison of the track ensembles indicates a number of systems with a broad range of intensities that compare well among the reanalyses. In addition, a number of small-scale weak systems are found that have no correspondence among the reanalyses or that only correspond upon relaxing the matching criteria, indicating possible differences in location and/or temporal coherence. These are distributed throughout the storm tracks, particularly in the regions known for small-scale activity, such as secondary development regions and the Mediterranean. For the Southern Hemisphere (SH), agreement is found to be generally less consistent in the lower troposphere with significant differences in both track density and mean intensity. The systems that correspond between the various reanalyses are considerably reduced and those that do not match span a broad range of storm intensities. Relaxing the matching criteria indicates that there is a larger degree of uncertainty in both the location of systems and their intensities compared with the NH. At upper-tropospheric levels, significant differences in the level of activity occur between the ECMWF reanalysis and the other reanalyses in both the NH and SH winters. This occurs due to a lack of coherence in the apparent propagation of the systems in ERA15 and appears most acute above 500 hPa. This is probably due to the use of optimal interpolation data assimilation in ERA15. Also shown are results based on using the same techniques to diagnose the tropical easterly wave activity. Results indicate that the wave activity is sensitive not only to the resolution and assimilation methods used but also to the model formulation.

Stereo correspondence estimation using multiwavelets scale-space representation-based multiresolution analysis

A multiresolution technique based on multiwavelets scale-space representation for stereo correspondence estimation is presented. The technique uses the well-known coarse-to-fine strategy, involving the calculation of stereo correspondences at the coarsest resolution level with consequent refinement up to the finest level. Vector coefficients of the multiwavelets transform modulus are used as corresponding features, where modulus maxima defines the shift invariant high-level features (multiscale edges) with phase pointing to the normal of the feature surface. The technique addresses the estimation of optimal corresponding points and the corresponding 2D disparity maps. Illuminative variation that can exist between the perspective views of the same scene is controlled using scale normalization at each decomposition level by dividing the details space coefficients with approximation space. The problems of ambiguity, explicitly, and occlusion, implicitly, are addressed by using a geometric topological refinement procedure. Geometric refinement is based on a symbolic tagging procedure introduced to keep only the most consistent matches in consideration. Symbolic tagging is performed based on probability of occurrence and multiple thresholds. The whole procedure is constrained by the uniqueness and continuity of the corresponding stereo features. The comparative performance of the proposed algorithm with eight famous existing algorithms, presented in the literature, is shown to validate the claims of promising performance of the proposed algorithm.

Concept of Matching Parallelepiped and its use in the correspondence problem

In this paper, the concept of Matching Parallelepiped (MP) is presented. It is shown that the volume of the MP can be used as an additional measure of `distance' between a pair of candidate points in a matching algorithm by Relaxation Labeling (RL). The volume of the MP is related with the Epipolar Geometry and the use of this measure works as an epipolar constraint in a RL process, decreasing the efforts in the matching algorithm since it is not necessary to explicitly determine the equations of the epipolar lines and to compute the distance of a candidate point to each epipolar line. As at the beginning of the process the Relative Orientation (RO) parameters are unknown, a initial matching based on gradient, intensities and correlation is obtained. Based on this set of labeled points the RO is determined and the epipolar constraint included in the algorithm. The obtained results shown that the proposed approach is suitable to determine feature-point matching with simultaneous estimation of camera orientation parameters even for the cases where the pair of optical axes are not parallel.

An Empirical Study of Non-Rigid Surface Feature Matching of Human from 3D Video

This paper presents an empirical study of affine invariant feature detectors to perform matching on video sequences of people with non-rigid surface deformation. Recent advances in feature detection and wide baseline matching have focused on static scenes. Video frames of human movement capture highly non-rigid deformation such as loose hair, cloth creases, skin stretching and free flowing clothing. This study evaluates the performance of six widely used feature detectors for sparse temporal correspondence on single view and multiple view video sequences. Quantitative evaluation is performed of both the number of features detected and their temporal matching against and without ground truth correspondence. Recall-accuracy analysis of feature matching is reported for temporal correspondence on single view and multiple view sequences of people with variation in clothing and movement. This analysis identifies that existing feature detection and matching algorithms are unreliable for fast movement with common clothing.

Feature LDA:a supervised topic model for automatic detection of web API documentations from the web

Web APIs have gained increasing popularity in recent Web service technology development owing to its simplicity of technology stack and the proliferation of mashups. However, efficiently discovering Web APIs and the relevant documentations on the Web is still a challenging task even with the best resources available on the Web. In this paper we cast the problem of detecting the Web API documentations as a text classification problem of classifying a given Web page as Web API associated or not. We propose a supervised generative topic model called feature latent Dirichlet allocation (feaLDA) which offers a generic probabilistic framework for automatic detection of Web APIs. feaLDA not only captures the correspondence between data and the associated class labels, but also provides a mechanism for incorporating side information such as labelled features automatically learned from data that can effectively help improving classification performance. Extensive experiments on our Web APIs documentation dataset shows that the feaLDA model outperforms three strong supervised baselines including naive Bayes, support vector machines, and the maximum entropy model, by over 3% in classification accuracy. In addition, feaLDA also gives superior performance when compared against other existing supervised topic models.