Spatio temporal feature evaluation for action recognition

Spatio-Temporal interest points are the most popular feature representation in the field of action recognition. A variety of methods have been proposed to detect and describe local patches in video with several techniques reporting state of the art performance for action recognition. However, the reported results are obtained under different experimental settings with different datasets, making it difficult to compare the various approaches. As a result of this, we seek to comprehensively evaluate state of the art spatio- temporal features under a common evaluation framework with popular benchmark datasets (KTH, Weizmann) and more challenging datasets such as Hollywood2. The purpose of this work is to provide guidance for researchers, when selecting features for different applications with different environmental conditions. In this work we evaluate four popular descriptors (HOG, HOF, HOG/HOF, HOG3D) using a popular bag of visual features representation, and Support Vector Machines (SVM)for classification. Moreover, we provide an in-depth analysis of local feature descriptors and optimize the codebook sizes for different datasets with different descriptors. In this paper, we demonstrate that motion based features offer better performance than those that rely solely on spatial information, while features that combine both types of data are more consistent across a variety of conditions, but typically require a larger codebook for optimal performance.

miRDeep*: an integrated application tool for miRNA identification from RNA sequencing data

miRDeep and its varieties are widely used to quantify known and novel micro RNA (miRNA) from small RNA sequencing (RNAseq). This article describes miRDeep*, our integrated miRNA identification tool, which is modeled off miRDeep, but the precision of detecting novel miRNAs is improved by introducing new strategies to identify precursor miRNAs. miRDeep* has a user-friendly graphic interface and accepts raw data in FastQ and Sequence Alignment Map (SAM) or the binary equivalent (BAM) format. Known and novel miRNA expression levels, as measured by the number of reads, are displayed in an interface, which shows each RNAseq read relative to the pre-miRNA hairpin. The secondary pre-miRNA structure and read locations for each predicted miRNA are shown and kept in a separate figure file. Moreover, the target genes of known and novel miRNAs are predicted using the TargetScan algorithm, and the targets are ranked according to the confidence score. miRDeep* is an integrated standalone application where sequence alignment, pre-miRNA secondary structure calculation and graphical display are purely Java coded. This application tool can be executed using a normal personal computer with 1.5 GB of memory. Further, we show that miRDeep* outperformed existing miRNA prediction tools using our LNCaP and other small RNAseq datasets. miRDeep* is freely available online at http://www.australianprostatecentre.org/research/software/mirdeep-star

Kernel analysis on Grassmann manifolds for action recognition

Modelling video sequences by subspaces has recently shown promise for recognising human actions. Subspaces are able to accommodate the effects of various image variations and can capture the dynamic properties of actions. Subspaces form a non-Euclidean and curved Riemannian manifold known as a Grassmann manifold. Inference on manifold spaces usually is achieved by embedding the manifolds in higher dimensional Euclidean spaces. In this paper, we instead propose to embed the Grassmann manifolds into reproducing kernel Hilbert spaces and then tackle the problem of discriminant analysis on such manifolds. To achieve efficient machinery, we propose graph-based local discriminant analysis that utilises within-class and between-class similarity graphs to characterise intra-class compactness and inter-class separability, respectively. Experiments on KTH, UCF Sports, and Ballet datasets show that the proposed approach obtains marked improvements in discrimination accuracy in comparison to several state-of-the-art methods, such as the kernel version of affine hull image-set distance, tensor canonical correlation analysis, spatial-temporal words and hierarchy of discriminative space-time neighbourhood features.

Improved foreground detection via block-based classifier cascade with probabilistic decision integration

Background subtraction is a fundamental low-level processing task in numerous computer vision applications. The vast majority of algorithms process images on a pixel-by-pixel basis, where an independent decision is made for each pixel. A general limitation of such processing is that rich contextual information is not taken into account. We propose a block-based method capable of dealing with noise, illumination variations, and dynamic backgrounds, while still obtaining smooth contours of foreground objects. Specifically, image sequences are analyzed on an overlapping block-by-block basis. A low-dimensional texture descriptor obtained from each block is passed through an adaptive classifier cascade, where each stage handles a distinct problem. A probabilistic foreground mask generation approach then exploits block overlaps to integrate interim block-level decisions into final pixel-level foreground segmentation. Unlike many pixel-based methods, ad-hoc postprocessing of foreground masks is not required. Experiments on the difficult Wallflower and I2R datasets show that the proposed approach obtains on average better results (both qualitatively and quantitatively) than several prominent methods. We furthermore propose the use of tracking performance as an unbiased approach for assessing the practical usefulness of foreground segmentation methods, and show that the proposed approach leads to considerable improvements in tracking accuracy on the CAVIAR dataset.

Research data finder : an institutional metadata repository

QUT’s new metadata repository (data registry), Research Data Finder, has been designed to promote the visibility and discoverability of QUT research datasets. Funded by the Australian National Data Service (ANDS), it will provide a qualitative snapshot of research data outputs created or collected by members of the QUT research community that are available via open or mediated access. As a fully integrated metadata repository Research Data Finder aligns with institutional sources of truth, such as QUT’s research administrative system, ResearchMaster, as well as QUT’s Academic Profiles system to provide high quality data descriptions that increase awareness of, and access to, shareable research data. In addition, the repository and its workflows are designed to foster smoother data management practices, enhance opportunities for collaboration and research, promote cross-disciplinary research and maximize existing research datasets. The metadata schema used in Research Data Finder is the Registry Interchange Format - Collections and Services (RIF-CS), developed by ANDS in 2009. This comprehensive schema is potentially complex for researchers; unlike metadata for publications, which are often made publicly available with the official publication, metadata for datasets are not typically available and need to be created. Research Data Finder uses a hybrid self-deposit and mediated deposit system. In addition to automated ingests from ResearchMaster (research project information) and Academic Profiles system (researcher information), shareable data is identified at a number of key “trigger points” in the research cycle. These include: research grant proposals; ethics applications; Data Management Plans; Liaison Librarian data interviews; and thesis submissions. These ingested records can be supplemented with related metadata including links to related publications, such as those in QUT ePrints. Records deposited in Research Data Finder are harvested by ANDS and made available to a national and international audience via Research Data Australia, ANDS’ discovery service for Australian research data. Researcher and research group metadata records are also harvested by the National Library of Australia (NLA) and these records are then published in Trove (the NLA’s digital information portal). By contributing records to the national infrastructure, QUT data will become more visible. Within Australia and internationally, many funding bodies have already mandated the open access of publications produced from publicly funded research projects, such as those supported by the Australian Research Council (ARC), or the National Health and Medical Research Council (NHMRC). QUT will be well placed to respond to the rapidly evolving climate of research data management. This project is supported by the Australian National Data Service (ANDS). ANDS is supported by the Australian Government through the National Collaborative Research Infrastructure Strategy Program and the Education Investment Fund (EIF) Super Science Initiative.

OpenRatSLAM : an open source brain-based SLAM system

RatSLAM is a navigation system based on the neural processes underlying navigation in the rodent brain, capable of operating with low resolution monocular image data. Seminal experiments using RatSLAM include mapping an entire suburb with a web camera and a long term robot delivery trial. This paper describes OpenRatSLAM, an open-source version of RatSLAM with bindings to the Robot Operating System framework to leverage advantages such as robot and sensor abstraction, networking, data playback, and visualization. OpenRatSLAM comprises connected ROS nodes to represent RatSLAM’s pose cells, experience map, and local view cells, as well as a fourth node that provides visual odometry estimates. The nodes are described with reference to the RatSLAM model and salient details of the ROS implementation such as topics, messages, parameters, class diagrams, sequence diagrams, and parameter tuning strategies. The performance of the system is demonstrated on three publicly available open-source datasets.

On robust face recognition via sparse encoding : the good, the bad, and the ugly

In the field of face recognition, Sparse Representation (SR) has received considerable attention during the past few years. Most of the relevant literature focuses on holistic descriptors in closed-set identification applications. The underlying assumption in SR-based methods is that each class in the gallery has sufficient samples and the query lies on the subspace spanned by the gallery of the same class. Unfortunately, such assumption is easily violated in the more challenging face verification scenario, where an algorithm is required to determine if two faces (where one or both have not been seen before) belong to the same person. In this paper, we first discuss why previous attempts with SR might not be applicable to verification problems. We then propose an alternative approach to face verification via SR. Specifically, we propose to use explicit SR encoding on local image patches rather than the entire face. The obtained sparse signals are pooled via averaging to form multiple region descriptors, which are then concatenated to form an overall face descriptor. Due to the deliberate loss spatial relations within each region (caused by averaging), the resulting descriptor is robust to misalignment & various image deformations. Within the proposed framework, we evaluate several SR encoding techniques: l1-minimisation, Sparse Autoencoder Neural Network (SANN), and an implicit probabilistic technique based on Gaussian Mixture Models. Thorough experiments on AR, FERET, exYaleB, BANCA and ChokePoint datasets show that the proposed local SR approach obtains considerably better and more robust performance than several previous state-of-the-art holistic SR methods, in both verification and closed-set identification problems. The experiments also show that l1-minimisation based encoding has a considerably higher computational than the other techniques, but leads to higher recognition rates.

Context-aware device self-configuration using self-organizing maps

Modern mobile computing devices are versatile, but bring the burden of constant settings adjustment according to the current conditions of the environment. While until today, this task has to be accomplished by the human user, the variety of sensors usually deployed in such a handset provides enough data for autonomous self-configuration by a learning, adaptive system. However, this data is not fully available at certain points in time, or can contain false values. Handling potentially incomplete sensor data to detect context changes without a semantic layer represents a scientific challenge which we address with our approach. A novel machine learning technique is presented - the Missing-Values-SOM - which solves this problem by predicting setting adjustments based on context information. Our method is centered around a self-organizing map, extending it to provide a means of handling missing values. We demonstrate the performance of our approach on mobile context snapshots, as well as on classical machine learning datasets.

Hot topic detection in news blogs from the perspective of W2T

News blog hot topics are important for the information recommendation service and marketing. However, information overload and personalized management make the information arrangement more difficult. Moreover, what influences the formation and development of blog hot topics is seldom paid attention to. In order to correctly detect news blog hot topics, the paper first analyzes the development of topics in a new perspective based on W2T (Wisdom Web of Things) methodology. Namely, the characteristics of blog users, context of topic propagation and information granularity are unified to analyze the related problems. Some factors such as the user behavior pattern, network opinion and opinion leader are subsequently identified to be important for the development of topics. Then the topic model based on the view of event reports is constructed. At last, hot topics are identified by the duration, topic novelty, degree of topic growth and degree of user attention. The experimental results show that the proposed method is feasible and effective.

A new database compilation of whole-rock chemical and geochronological data of igneous rocks in Queensland : a new resource for HDR geothermal resource exploration

The Geothermal industry in Australia and Queensland is in its infancy and for hot dry rock (HDR) geothermal energy, it is very much in the target identification and resource definition stages. As a key effort to assist the geothermal industry and exploration for HDR in Queensland, we are developing a comprehensive and new integrated geochemical and geochronological database on igneous rocks. To date, around 18,000 igneous rocks have been analysed across Queensland for chemical and/or age information. However, these data currently reside in a number of disparate datasets (e.g., Ozchron, Champion et al., 2007, Geological Survey of Queensland, journal publications, and unpublished university theses). The goal of this project is to collate and integrate these data on Queensland igneous rocks to improve our understanding of high heat producing granites in Queensland, in terms of their distribution (particularly in the subsurface), dimensions, ages, and controlling factors in their genesis.

An ensemble method for predicting subnuclear localizations from primary protein structures

Background Predicting protein subnuclear localization is a challenging problem. Some previous works based on non-sequence information including Gene Ontology annotations and kernel fusion have respective limitations. The aim of this work is twofold: one is to propose a novel individual feature extraction method; another is to develop an ensemble method to improve prediction performance using comprehensive information represented in the form of high dimensional feature vector obtained by 11 feature extraction methods. Methodology/Principal Findings A novel two-stage multiclass support vector machine is proposed to predict protein subnuclear localizations. It only considers those feature extraction methods based on amino acid classifications and physicochemical properties. In order to speed up our system, an automatic search method for the kernel parameter is used. The prediction performance of our method is evaluated on four datasets: Lei dataset, multi-localization dataset, SNL9 dataset and a new independent dataset. The overall accuracy of prediction for 6 localizations on Lei dataset is 75.2% and that for 9 localizations on SNL9 dataset is 72.1% in the leave-one-out cross validation, 71.7% for the multi-localization dataset and 69.8% for the new independent dataset, respectively. Comparisons with those existing methods show that our method performs better for both single-localization and multi-localization proteins and achieves more balanced sensitivities and specificities on large-size and small-size subcellular localizations. The overall accuracy improvements are 4.0% and 4.7% for single-localization proteins and 6.5% for multi-localization proteins. The reliability and stability of our classification model are further confirmed by permutation analysis. Conclusions It can be concluded that our method is effective and valuable for predicting protein subnuclear localizations. A web server has been designed to implement the proposed method. It is freely available at http://bioinformatics.awowshop.com/snlpr​ed_page.php.

Bus and car travel time on urban networks : integrating Bluetooth and bus vehicle identification data

Travel time in an important transport performance indicator. Different modes of transport (buses and cars) have different mechanical and operational characteristics, resulting in significantly different travel behaviours and complexities in multimodal travel time estimation on urban networks. This paper explores the relationship between bus and car travel time on urban networks by utilising the empirical Bluetooth and Bus Vehicle Identification data from Brisbane. The technologies and issues behind the two datasets are studied. After cleaning the data to remove outliers, the relationship between not-in-service bus and car travel time and the relationship between in-service bus and car travel time are discussed. The travel time estimation models reveal that the not-in-service bus travel time are similar to the car travel time and the in-service bus travel time could be used to estimate car travel time during off-peak hours

Application of multi-criteria decision making methods to compression ignition engine efficiency and gaseous, particulate and greenhouse gas emissions

Compression ignition (CI) engine design is subject to many constraints which presents a multi-criteria optimisation problem that the engine researcher must solve. In particular, the modern CI engine must not only be efficient, but must also deliver low gaseous, particulate and life cycle greenhouse gas emissions so that its impact on urban air quality, human health, and global warming are minimised. Consequently, this study undertakes a multi-criteria analysis which seeks to identify alternative fuels, injection technologies and combustion strategies that could potentially satisfy these CI engine design constraints. Three datasets are analysed with the Preference Ranking Organization Method for Enrichment Evaluations and Geometrical Analysis for Interactive Aid (PROMETHEE-GAIA) algorithm to explore the impact of 1): an ethanol fumigation system, 2): alternative fuels (20 % biodiesel and synthetic diesel) and alternative injection technologies (mechanical direct injection and common rail injection), and 3): various biodiesel fuels made from 3 feedstocks (i.e. soy, tallow, and canola) tested at several blend percentages (20-100 %) on the resulting emissions and efficiency profile of the various test engines. The results show that moderate ethanol substitutions (~20 % by energy) at moderate load, high percentage soy blends (60-100 %), and alternative fuels (biodiesel and synthetic diesel) provide an efficiency and emissions profile that yields the most “preferred” solutions to this multi-criteria engine design problem. Further research is, however, required to reduce Reactive Oxygen Species (ROS) emissions with alternative fuels, and to deliver technologies that do not significantly reduce the median diameter of particle emissions.

Use of concept maps in reforming teaching Family and Consumer Sciences in Kuwait

This paper addresses the professional development of Kuwaiti teachers in the use of concept maps to teach Family and Consumer Science. A key aim of the study was to evaluate the degree to which the use of concept maps would influence the way Kuwaiti teachers approach and teach Family and Consumer Studies (FCS) subjects and the degree to which concept maps empower students to critically identify and express their knowledge of the subject being taught. A case study methodology was adopted to follow the implementation of lessons using concept maps by four teachers of middle years. An analysis of the data revealed the positive impact that student-centred teaching tools can have on the reformation of traditional teaching environments. For all teachers, the primary strengths of using concept maps were the ability to generate student interest, to motivate student participation and to enhance student understanding of content. Although a case study design may limit the generalisation and comparative value of the study, the findings of this study remain important to the planning of future professional development programs and the use of concept maps within Kuwait’s FCS curriculum area.

Comparing child product safety concerns with injury incidents : does the evidence support the response?

Background Efficient effective child product safety (PS) responses require data on hazards, injury severity and injury probability. PS responses in Australia largely rely on reports from manufacturers/retailers, other jurisdictions/regulators, or consumers. The extent to which reactive responses reflect actual child injury priorities is unknown. Aims/Objectives/Purpose This research compared PS issues for children identified using data compiled from PS regulatory data and data compiled from health data sources in Queensland, Australia. Methods PS regulatory documents describing issues affecting children in Queensland in 2008–2009 were compiled and analysed to identify frequent products and hazards. Three health data sources (ED, injury surveillance and hospital data) were analysed to identify frequent products and hazards. Results/Outcomes Projectile toys/squeeze toys were the priority products for PS regulators with these toys having the potential to release small parts presenting choking hazards. However, across all health datasets, falls were the most common mechanism of injury, and several of the products identified were not subject to a PS system response. While some incidents may not require a response, a manual review of injury description text identified child poisonings and burns as common mechanisms of injuries in the health data where there was substantial documentation of product-involvement, yet only 10% of PS system responses focused on these two mechanisms combined. Significance/contribution to the field Regulatory data focused on products that fail compliance checks with ‘potential’ to cause harm, and health data identified actual harm, resulting in different prioritisation of products/mechanisms. Work is needed to better integrate health data into PS responses in Australia.