Thursdays, 12-1 pm, WS 318 (Boardroom of AUT Engineering)
- 10 December 2015
A/Prof. Wei Qi Yan (AUT, SECMS): Face Search in Encrypted Domain
Visual information of images and videos usually is encrypted for the purposes of security applications. Straightforward manipulations on the encrypted data without requiring any decryption have the advantage of speed over performing those operations in spatial, temporal, frequency or compressed domain. In this talk, we will present encrypted image search. More specifically, given a face image as the target object, we search it amongst encrypted images. We accomplish the search by using a novel method that extracts features and locates the face object region within the given encrypted image. We evaluate the search results by using precision and recall as well as F-measure. Our experiments reveal that there exists a trade-off between the quality of search and the quality of encryption, namely, stronger encryption leads to poorer search performance.
- 4 December 2015 (Friday)
Prof. Pedro Real (University of Seville, Spain): A Strategy for Parallelizing Topological Computation within Digital Images — Homological Spanning Forest
In this talk, we discuss a recent mathematical model developing a framework suitable for parallel topological computation for digital images. More precisely, it is a purely combinatorial version of the algebraic-topological model called “Homological Spanning Forest”. This framework provides topological coordinates for the different cells of an abstract cell complex structure K(O) associated to a digital object O. In order to take advantage of the inherent massive data parallelism that any image processing has, the fundamental topological tools have been transformed so as to promote an efficient parallel implementation in any parallel-oriented architecture (GPUs, multithreaded computers, SIMD kernels and so on). In particular, iterative and conditional sentences have been almost eliminated, any complex data structures have been avoided in favour of matrix descriptions, and most of processing has been reduced to element-by-element operations. A software prototype modelling such framework is developed, which demonstrates the strength of the parallel framework. Finally, some experiments for computing topological information related to homology and cohomology with coefficients in a field and the fundamental group of a digital object validates the approach.
- 23-27 November: IVCNZ and PSIVT 2015 at Auckland
- 19 November 2015
Matthias Ochs (Goethe University Frankfurt-Main, Germany): Computer Vision Research for Traffic and Autonomous Driving Applications
Driver Assistance Systems and Autonomous Driving currently belong to the most demanding industrial application fields for Computer Vision. At the VSI lab in Frankfurt, Germany, these kinds of applications are in the focus of several research projects that aim at creating new algorithms for analysing data from monocular and multiple cameras. This talk will present several new VSI developments in the area of motion processing. One of the fundamental techniques is a new enhanced version of the classical phase correlation approach which has the capability of computing a coarse overall description of the motion inside an observation window. This method is used in several different contexts, e.g. for guiding matches based on epipolar information, and also for computing the rotational part of egomotion before a detailed motion field is computed. We build strongly on dynamical models for egomotion, as the movement of vehicles is usually strongly constrained and highly predictable.
- 12 November 2015
Dr. Shahper Vodanovich (AUT, SECMS): Digital Natives
The term ‘Digital Natives’ commonly refers to young people who have been born and raised in a world where technologies are ubiquitous. Digital natives are meeting the common challenges of adolescence such as the development of social, cognitive and emotional well-being in these ubiquitous spaces. While research on ubiquitous spaces within the Information Systems domain is gaining popularity, there is very little emphasis on the role of digital natives within these spaces. Similarly, digital native use of the Internet as a ubiquitous space is a popular research area within the youth development and educational domains. However, despite the popular interest in this topic there is insufficient research on understanding how these ubiquitous spaces are impacting on the well-being and development of digital natives. One of the key objectives of this research was to explore digital native use of ubiquitous spaces and to provide a coherent understanding of their well-being and development in these spaces. Subsequently, the second key objective was to explain factors that may influence the impact of ubiquitous spaces on digital native well-being and development. The findings suggest that digital natives with personality type extravert, agreeableness or neuroticism significantly moderate the relationship between use of SNTs and social competence. In addition, contrary to existing research, the data revealed that gender does not moderate this relationship. The proposed models and frameworks enable a coherent understanding of the well-being and development of digital natives in ubiquitous spaces.
- 05 November 2015 (WT 515C)
Dr. Boris Bacic (AUT, SECMS): Algorithmic and third generation ANN approaches for temporal and spatial pattern recognition: A case study on tennis
Temporal and spatial event recognition methods have broad multi-disciplinary applications in the areas of speech recognition, surveillance, exergaming, biomechanics, rehabilitation and sports science in general.
Common approaches include various algorithmic and artificial neural approaches such as sliding-window, feature extraction and pattern recognition. Three-dimensional tennis motion data have been captured (as 22-marker time series at 50 fps) using fixed-location multi-camera video of one expert performing ground strokes mimicking various good swings and common errors that are typical for novice to intermediate players.
Presented solutions on forehand/backhand classification and swing stance angle extraction are based on expert insights and algorithmic implementation of common-sense coaching rules. In addition, the presented swing event extraction model is based on third-generation artificial neural networks relying on learning-by-example from data and did not require specific insights on data or knowledge of common-sense rules. The presented solutions have achieved 100% event detection compared to experts and are applicable to multiple areas such as: improving safety in surveillance; robot and automotive vision; automation of mutli-time series data analysis; augmented coaching technology; rehabilitation and exergaming systems.
- End-of-semester break
- 22 October 2015
Yanchang Xiao (Computer Science, CJLU) Building Detection in Natural Scenes Based on DRF
Based on the DRFs framework, in this talk we firstly simplify the interaction potential and the associated potential model, we reduce dimension of the multi-scale features, re-define dimension of the single-scale feature, and increase the colour feature of the target building object. Furthermore, the quasi-Newton method combing with linear search and gradient descent method are adopted to find the parameters, we therefore get a simple model and achieve better performance. Finally, the partition function of the DRF is eliminated by using pseudo-likelihood method for parameter estimation. The simulation results show false positive of the proposed method is lower than that of Kumar and Hebert method, while the correct rate and detection rate are higher after these improvements.
- 15 October 2015
A/Prof. Russel Pears (AUT, SECMS): Data Stream Classification: Challenges, Applications and Solutions
In this talk the challenges involved in pattern recognition and knowledge extraction from a stream of continuously arriving data will be introduced. Data streams presents unique challenges that need to be overcome through the use of novel methods. Some of past research in this area that includes automatic methods of detecting when significant changes in patterns take place in a stream as well as pattern capture through the use of the Discrete Fourier Transform will be presented. Finally open research issues in the data stream area in relation to some important applications will be discussed.
- 8 October 2015
Dr. Barbara Breen (AUT, School of Applied Sciences): Video Data Collection by UAVs for Environmental SurveillanceDriver Assistance Systems and Autonomous Driving currently belong to the most demanding industrial application fields for Computer Vision. At the VSI lab in Frankfurt, Germany, these kinds of applications are in the focus of several research projects that aim at creating new algorithms for analyzing data from monocular and multiple cameras. The talk will present several new VSI developments in the area of motion processing. One of the fundamental techniques is a new enhanced version of the classical phase correlation approach which has the capability of computing a coarse overall description of the motion inside an observation window. This method is used in several different contexts, e.g. for guiding matches based on epipolar information, and also for computing the rotational part of egomotion before a detailed motion field is computed. We build strongly on dynamical models for egomotion, as the movement of vehicles is usually strongly constrained and highly predictable.
- 1 October 2015
Prof. Reinhard Klette (AUT, EEE): A few academic impressions during a current trip to World Congress of Robotics, CAIP and three universities at Wuhan
This talk reports about two conferences, the 1st World Congress of Robotics at Shenyang, China, and the 16th CAIP (Computer Analysis of Images and Patterns) conference at Valetta, Malta. CeRV had major activities at both conferences. The talk also reports about visits at three universities at Wuhan China: of NERC at Central China Normal University, of the School of Automation at Huazhong University of Science & Technology, and of institutes at Wuhan University, including the International School of Software, the Computer School, and the State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing. The talk also informs about research on vision-augmented vehicles at the Electronic Information School of Wuhan University.
- 24 September 2015
Dr. Junli Tao (Computer Science, UoA) and Rafael Guillermo (Mexico, Centro de investigaciones en Óptica): Registration and Segmentation of Aerial Images
- Mid-semester break
- 13 August 2015
Dr. Manoranjan Mohanty (Computer Science, UoA): Secret Sharing Approach for Securing Cloud-Based Image Processing
Advances in cloud computing have allowed image processing tasks to be outsourced to cloud datacentres. Although such cloud-based image processing have many advantages, security and privacy are two main concerns. Addressing these two concerns is challenging, since there is no fully homomorphic practical cryptosystem for floating point numbers. In this talk, we will discuss our approach of using Shamir’s secret sharing for securing two cloud-based image processing tasks, cloud-based image scaling/cropping and cloud-based volume ray-casting. Our scheme allows image scaling/cropping and volume ray-casting operations to be outsourced to a cloud datacentre without revealing the content of image or data to it. In addition to providing confidentiality and integrity, our scheme incurs acceptable overheads.
- 6 August 2015
A/Prof. Wei Qi Yan(AUT, SECMS): Privacy Preservation of Social Media
Privacy preservation on social networks is a leading-edge research problem in this era. Despite the research work has achieved remarkable progression in preserving the privacy of user profiles of social networks and visual surveillance, the existing problems in social media have not taken a gratified step. The reason is for the lack of effective modelling, computational algorithms and resultant evaluations in quantitative research. In this talk, we would like to take social media into consideration and link users together under the umbrella of social networks so as to exploit a way that the existing problems related to media privacy could be resolved. Our results attest the proposed privacy persevering approaches are effective and have excellent performance in media privacy preservation.
- 30 July 2015 (12-1pm, WS102)
Jonas Hilty (AUT, School of Applied Sciences): Three-dimensional vegetation modelling using unmanned aerial vehicle derived photographs
The application of Structure from Motion algorithms to photographs derived from unmanned aerial vehicles equipped with consumer grade cameras is a cheap method of three-dimensional vegetation modelling. This talk gives an overview of current application of this technology in forest ecology.
- 23 July 2015
AUT Showcase (no seminar)
- 16 July 2015
Rafael Guillermo (Mexico, Centro de investigaciones en Óptica): Explanation of the Main Algorithms in Particle Image Velocimetry
This talk explains the basic idea of particle image velocimetry. The correlation algorithm and the Lucas Kanade algorithm are both analysed, step by step, with mathematical proofs for each step. It is also explained how applications of Artificial Intelligence is integrated into the Lucas Kanade algorthim.
- 9 July 2015
Prof. Reinhard Klette (AUT, EEE): Three Novel Hough Transform Techniques
The talk informs about three different and novel approaches for detecting linear patterns in images, a statistical method, an entropy-based method, and a closed form solution. Results have been currently published in journals Pattern Recognition, Computer Vision and Image Understanding, and IEEE Transactions on Image Processing. This has been joint work with Prof. Zezhong Xu and Dr. Bok-Suk Shin.
- 4 June 2015
Haokun Geng (UoA, CS), Ron Nair (AUT), and Harshpreet Singh (AUT): New CeRV website – Discussion of two options
- 28 May 2015
Robert Yang (AUT, EEE): The Use of Video to Detect and Measure Pollen on Bees Entering a Beehive
Bee populations have been observed for a long time. The health of bees affects the population. There are many factors that affect bees’ health. One factor is nutrition. Bees need an adequate quantity and different varieties of pollen. This research will measure the pollen being brought into the beehive in the bees’ pollen sacs. It will process 2-D video which is obtained from in front of the entrance to the beehive and count the number of full pollen sacs which bees bring back.
- 21 May 2015
Dr. Boris Bacic (AUT, Computer Science):
Development of Augmented Coaching Systems (ACS): Combining Matlab and Delphi for implementation of Personal Tennis Coaching System The application “Personal Tennis Coaching System” utilises captured 3D motion data of tennis activity, without information about the ball impact. The purpose of the ACS – Personal Tennis Coaching System is to help end-users with motor learning and stylistic execution of tennis swings. Utilised AI approaches were implemented in Matlab. Interactive animated 3D player and the user interface were implemented in Delphi/Object Pascal codebase including presentation logic and session control with Matlab server. Future work is aimed at development of ACS for sport, human performance, injury prevention and rehabilitation. Future ACS should be able to process multi-modal motion data such as video, image, sound and other signals in stand-alone mode or distributed scenario
- 7 May 2015
Dr. John Robertson (AUT, Applied Sciences): Cheap Autonomous Vehicles Using Off-the-shelf Hobby Components
- 30 April 2015
Dr. Minh Nguyen (School of Computer and Mathematical Sciences, AUT): A Guided Dynamic Programming Stereo Algorithm
We investigate whether it is possible to combine several stereo matching algorithms in order to create a hybrid algorithm that is superior to its components. We introduce a novel variant of a dynamic programming stereo algorithm that allows for guidance by other algorithms. We make use of our desirability measure to experimentally demonstrate that a combination of stereo matching algorithms is able to produce new powerful hybrids.
- 23 April 2015
Junli Tao (Computer Science, UoA): Multi-target Tracking
This presentation reports about proposed tracking algorithms for various targets, such as Drosophila Melanogaster, vehicles, or pedestrians. The talk highlights specific and general technologies to be applied for multi-target tracking. This talk summarises work done in my PhD project.
- 26 March 2015
Haokun Geng (Computer Science, UoA): Visual odometry and scene reconstruction with Kalman filters and multi-sensor integration
Visual odometry is the process of estimating the actual vehicle’s (or cameras’) motion over stereo images. For our dynamic and complex environment on the road, only use the image data from optical cameras may not be sufficient in some extreme cases (i.e. closely following a big truck). So we use the multi-sensor integration to improve the reliability of our egomotion estimation methods. This talk introduces an approach for egomotion estimation over stereo images combined with extra GPS data. The cameras’ trajectory and 3D roadside reconstruction is used here as an evaluation method for the egomotion estimation.
- 19 March 2015
Dongwei Liu (Computer Science, UoA): Stereo refinement for photo editing
The talk introduces into the applied joint bilateral filter and two methods for refining depth maps obtained by stereo matching. Based on experimental results there will be comparisons and a discussion. The illustrated applications are about the generation of fog in stereo images, and also of depth-accurate bokeh.
- 26 February 2015
Bradley Moorfield: Refinement of 3D Roadside Shape Reconstructions
(Report about a Summer Research Project)
- 5 February 2015
Ali Al-Sarraf and Kang Kong: Final reports about MSc projects
- 22 January 2015
Haokun Geng: Egomotion estimation using additional data sources