Brian Lovell | The University of Queensland, Australia (original) (raw)

Papers by Brian Lovell

Proceedings of the 17th International Conference on Pattern Recognition, 2004. ICPR 2004., 2004

We propose a novel energy minimisation framework for the dense reconstruction of stereo image seq... more We propose a novel energy minimisation framework for the dense reconstruction of stereo image sequences that incorporates data fidelity as well as spatial and temporal regularity. An iterated dynamic programming scheme is proposed to minimise the energy function. We also present an efficient implementation of the minimisation scheme by introducing morphological decomposition techniques to solve the dynamic programming subproblem. Our proposed method is capable of reconstructing dynamic scenes with complex motion. Results are presented demonstrating the strength of our proposed algorithm.

Handbook of Pattern Recognition and Computer Vision, 2005

This paper considers the recovery of the epipolar geometry for the case of calibrated/uncalibrate... more This paper considers the recovery of the epipolar geometry for the case of calibrated/uncalibrated two view relations. Two view relations are the basis for accurate calibration of stereo rigs and are often used in the solution to sequential processing of digital image streams to recover the surface structure of a scene or otherwise known as, structure from motion (SFM). The outcomes of this research are a freely available implementation of the algorithms required to determine an accurate solution to the epipolar geometry of a two-view relation.

Proceedings of the 17th International Conference on Pattern Recognition, 2004. ICPR 2004., 2004

A method for the classification of land cover in urban areas by the fusion of first and last puls... more A method for the classification of land cover in urban areas by the fusion of first and last pulse LIDAR data and multi-spectral images is presented. Apart from buildings, the classes "tree", "grass land", and "bare soil" are also distinguished by a classification method based on the theory of Dempster -Shafer for data fusion. Examples are given for a test site in Germany.

2012 International Conference on Digital Image Computing Techniques and Applications (DICTA), 2012

Person re-identification on image sets in which each image is taken from a different angle and li... more Person re-identification on image sets in which each image is taken from a different angle and lighting condition is a very challenging task. This task becomes even more difficult when images are low resolution and carrying image compression artifacts. The accuracy of the existing re-identification techniques are relatively low on the challenging evaluation grounds such as VIPeR and iLIDS image datasets. In these datasets, distortions in shape and colour make the re-identification task difficult and uncertain for both machine and human. In this paper, we propose a new approach to address the uncertainty in low resolution images for person re-identification by using statistical approximation. We first show that the distribution within a patch on person's image does not fit a normal distribution via Kolmogorov-Smirnov test. Then we simplify the Kolmogorov-Smirnov statistic by using only the mean and standard deviation of the distribution. These values are used as descriptors for per region per channel, and concatenated for comparison of image pairs. Experiments show that the proposed approach outperforms the state-of-the-art on person re-identification methods. The small memory foot print and the low computational cost of the proposed technique make it suitable for person re-identification in large scale surveillance applications.

2011 Fifth ACM/IEEE International Conference on Distributed Smart Cameras, 2011

We propose two novel techniques for automatic summarisation of lengthy surveillance videos, based... more We propose two novel techniques for automatic summarisation of lengthy surveillance videos, based on selection of frames containing scenes most informative for rapid perusal and interpretation by humans. In contrast to other video summarisation methods, the proposed methods explicitly focus on foreground objects, via edge histogram descriptor and a localised foreground information quantity (entropy) measurement. Frames are iteratively pruned until a preset summarisation rate is reached. Experiments on the publicly available CAVIAR dataset, as well as our own dataset focused on people walking through natural choke points (such as doors), suggest that the proposed method obtains considerably better results than methods based on optical flow, entropy differences and colour spatial distribution characteristics.

2009 Digital Image Computing: Techniques and Applications, 2009

ABSTRACT The inherent nature of image and video and its multi-dimension data space makes its proc... more ABSTRACT The inherent nature of image and video and its multi-dimension data space makes its processing and interpretation a very complex task, normally requiring considerable processing power. Moreover, understanding the meaning of video content and storing it in a fast searchable and readable form, requires taking advantage of image processing methods, which when running them on a video stream per query, would not be cost effective, and in some cases is quite impossible due to time restrictions. Hence, to speed up the search process, storing video and its extracted meta-data together is desired. The storage model itself is one of the challenges in this context, as based on the current CCTV technology; it is estimated to require a petabyte size data management system. This estimate however, is expected to grow rapidly as current advances in video recording devices are leading to higher resolution sensors, and larger frame size. On the other hand, the increasing demand for object tracking on video streams has invoked the research on content-based image retrieval (CBIR) and content-based video retrieval (CBVR). In this paper, we present the design and implementation of a framework and a data model for CCTV surveillance videos on RDBMS which provides the functions of a surveillance monitoring system, with a tagging structure for event detection. On account of some recent results, we believe this is a promising direction for surveillance video search in comparison to the existing solutions.

18th International Conference on Pattern Recognition (ICPR'06), 2006

Extracting human postural information from video sequences has proved a difficult research questi... more Extracting human postural information from video sequences has proved a difficult research question. The most successful approaches to date have been based on particle filtering, whereby the underlying probability distribution is approximated by a set of particles. The shape of the underlying observational probability distribution plays a significant role in determining the success, both accuracy and efficiency, of any visual tracker. In this paper we compare approaches used by other authors and present a cost path approach which is commonly used in image segmentation problems, however is currently not widely used in tracking applications.

9th Biennial Conference of the Australian Pattern Recognition Society on Digital Image Computing Techniques and Applications (DICTA 2007), 2007

Smart Cameras are becoming more popular in Intelligent Surveillance Systems area. Recognizing fac... more Smart Cameras are becoming more popular in Intelligent Surveillance Systems area. Recognizing faces in a crowd in real-time is a key features which would significantly enhance Intelligent Surveillance Systems. Using a high resolution smart camera as a tool to extract faces that are suitable for face recognition would greatly reduce the computational load on the main processing unit. This processing unit would not be overloaded by the demands of the high data rates required for high resolution video and could be designed solely for face recognition. In this paper we report on a multiple-stage face detection and tracking system that is designed for implementation on the NICTA high resolution (5 MP) smart camera.

2010 11th International Conference on Control Automation Robotics & Vision, 2010

This paper presents a novel generic weak classifier for object detection called &... more This paper presents a novel generic weak classifier for object detection called "Square Patch Feature". The speed and overall performance of a detector utilising Square Patch features in comparison to other weak classifiers shows improvement. Each weak classifier is based on the difference between two or four fixed size square patches in an image. A pre-calculated representation of the image

Proceedings of the Fourth ACM/IEEE International Conference on Distributed Smart Cameras - ICDSC '10, 2010

... The University of Queensland 4072, QLD, Australia and National ICT Australia PO Box 6020, St ... more ... The University of Queensland 4072, QLD, Australia and National ICT Australia PO Box 6020, St Lucia 4067, QLD, Australia yasir@itee.uq ... 1 –6. [15] J. Cho, B. Benson, S. Mirzaei, and R. Kastner, “Parallelized architecture of multiple classifiers for face detection,” in Application ...

IEEE Transactions on Signal Processing, 1992

This paper defines the 3D reconstruction problem as the process of reconstructing a 3D scene from... more This paper defines the 3D reconstruction problem as the process of reconstructing a 3D scene from numerous 2D visual images of that scene. It is well known that this problem is ill-posed, and numerous constraints and assumptions are used in 3D reconstruction algorithms in order to reduce the solution space. Unfortunately, most constraints only work in a certain range of situations and often constraints are built into the most fundamental methods (e.g. Area Based Matching assumes that all the pixels in the window belong to the same object). This paper presents a novel formulation of the 3D reconstruction problem, using a voxel framework and first order logic equations, which does not contain any additional constraints or assumptions. Solving this formulation for a set of input images gives all the possible solutions for that set, rather than picking a solution that is deemed most likely. Using this formulation, this paper studies the problem of uniqueness in 3D reconstruction and how the solution space changes for different configurations of input images. It is found that it is not possible to guarantee a unique solution, no matter how many images are taken of the scene, they're orientation or even how much colour variation is in the scene itself. Results of using the formulation to reconstruct a few small voxel spaces are also presented. They show that the number of solutions is extremely large for even very small voxel spaces (5x5 voxel space gives 10 to 10 7 solutions). This shows the need for constraints to reduce the solution space to a reasonable size. Finally, it is noted that because of the discrete nature of the formulation, the solution space size can be easily calculated, making the formulation a useful tool to numerically evaluate the usefulness of any constraints that are added.

Visual tracking of the human body has attracted increasing attention due to the potential to perf... more Visual tracking of the human body has attracted increasing attention due to the potential to perform high volume low cost analyses of motions in a wide range of applications, including sports training, rehabilitation and security. In this paper we present the development of a visual tracking module for a system aimed to be used as an autonomous instructional aid for amateur golfers. Postural information is captured visually and fused with information from a golf swing analyser mat and both visual and audio feedback given based on the golfers mistakes. Results from the visual tracking module are presented.

We propose what we believe is a new approach to 3D reconstruction through the design of a 3D voxe... more We propose what we believe is a new approach to 3D reconstruction through the design of a 3D voxel volume, such that all the image information and camera geometry are embedded into one feature space. By customising the volume to be suitable for segmentation, the key idea that we propose is the recovery of a 3D scene through the use of globally optimal geodesic active contours. We also present an extension to this idea by proposing the novel design of a 4D voxel volume to analyse the stereo motion problem in multi-view image sequences.

Circuits and Systems, …, May 8, 1989

A procedure is described for the time-frequency analysis of signals, based on time-frequency dist... more A procedure is described for the time-frequency analysis of signals, based on time-frequency distributions (TFDs) and instantaneous frequency (IF) estimation. First, a suitable TFD is used to determine the number of signal components. Then, if the signal is monocomponent, the IF law can be estimated directly. For multicomponent signals, two-dimensional windowing in the time-frequency domain (a form of time-varying filtering) is used to isolate each component; IF estimation is then applied to the individual ...

The Research Network for a Secure Australia (RNSA) is a multidisciplinary collaboration establish... more The Research Network for a Secure Australia (RNSA) is a multidisciplinary collaboration established to strengthen Australia's research capacity for protecting critical infrastructure (CIP) from natural or human caused disasters including terrorist acts. The RNSA facilitates a knowledge-sharing network for research organisations, government and the private sector to develop research tools and methods to mitigate emerging safety and security issues relating to critical infrastructure. World-leaders with extensive national and international linkages in relevant scientific, engineering and technological research will lead this collaboration. The RNSA also organises various activities to foster research collaboration and nurture young investigators.

The huge popularity of Hidden Markov models in pattern recognition is due to the ability to "lear... more The huge popularity of Hidden Markov models in pattern recognition is due to the ability to "learn" model parameters from an observation sequence through Baum-Welch and other re-estimation procedures. In the case of HMM parameter estimation from an ensemble of observation sequences, rather than a single sequence, we require techniques for finding the parameters which maximize the likelihood of the estimated model given the entire set of observation sequences. The importance of this study is that HMMs with parameters estimated from multiple observations are shown to be many orders of magnitude more probable than HMM models learned from any single observation sequence -thus the effectiveness of HMM "learning" is greatly enhanced. In this paper, we present techniques that usually find models significantly more likely than Rabiner's wellknown method on both seen and unseen sequences.

… , 2005. ITCC 2005. …, 2005

Frank Adelstein Dharma P. Agrawal Igor Aizenberg Giovanni Aloisio Kazumaro Aoki Hamid Arabnia Vij... more Frank Adelstein Dharma P. Agrawal Igor Aizenberg Giovanni Aloisio Kazumaro Aoki Hamid Arabnia Vijayan Asari Michail Attalah Robert L. Baber Pascal Bamford Nick Barnes Emad Bataineh Lejla Batina Siddika Berna Ors Guido Bertoni Euro Blasi Rainer Bluemel Luca Breveglieri Constantine Butakoff Greg Byrd Massimo Cafaro Miriam Capretz Gabriele Carteni Jordi Castella-Roca Herwin Chan Pei-Min Chen Alex Chen Chia-Chu Chiang Jagadish Chintala Edward Christensen Chi-Kit Ronald Chung Vaughan Clarkson Pedro Henrique G. Coelho Nedeljko Cvejic ...

Proceedings of the 17th International Conference on Pattern Recognition, 2004. ICPR 2004., 2004

We propose a novel energy minimisation framework for the dense reconstruction of stereo image seq... more We propose a novel energy minimisation framework for the dense reconstruction of stereo image sequences that incorporates data fidelity as well as spatial and temporal regularity. An iterated dynamic programming scheme is proposed to minimise the energy function. We also present an efficient implementation of the minimisation scheme by introducing morphological decomposition techniques to solve the dynamic programming subproblem. Our proposed method is capable of reconstructing dynamic scenes with complex motion. Results are presented demonstrating the strength of our proposed algorithm.

Handbook of Pattern Recognition and Computer Vision, 2005

This paper considers the recovery of the epipolar geometry for the case of calibrated/uncalibrate... more This paper considers the recovery of the epipolar geometry for the case of calibrated/uncalibrated two view relations. Two view relations are the basis for accurate calibration of stereo rigs and are often used in the solution to sequential processing of digital image streams to recover the surface structure of a scene or otherwise known as, structure from motion (SFM). The outcomes of this research are a freely available implementation of the algorithms required to determine an accurate solution to the epipolar geometry of a two-view relation.

Proceedings of the 17th International Conference on Pattern Recognition, 2004. ICPR 2004., 2004

A method for the classification of land cover in urban areas by the fusion of first and last puls... more A method for the classification of land cover in urban areas by the fusion of first and last pulse LIDAR data and multi-spectral images is presented. Apart from buildings, the classes "tree", "grass land", and "bare soil" are also distinguished by a classification method based on the theory of Dempster -Shafer for data fusion. Examples are given for a test site in Germany.

2012 International Conference on Digital Image Computing Techniques and Applications (DICTA), 2012

Person re-identification on image sets in which each image is taken from a different angle and li... more Person re-identification on image sets in which each image is taken from a different angle and lighting condition is a very challenging task. This task becomes even more difficult when images are low resolution and carrying image compression artifacts. The accuracy of the existing re-identification techniques are relatively low on the challenging evaluation grounds such as VIPeR and iLIDS image datasets. In these datasets, distortions in shape and colour make the re-identification task difficult and uncertain for both machine and human. In this paper, we propose a new approach to address the uncertainty in low resolution images for person re-identification by using statistical approximation. We first show that the distribution within a patch on person's image does not fit a normal distribution via Kolmogorov-Smirnov test. Then we simplify the Kolmogorov-Smirnov statistic by using only the mean and standard deviation of the distribution. These values are used as descriptors for per region per channel, and concatenated for comparison of image pairs. Experiments show that the proposed approach outperforms the state-of-the-art on person re-identification methods. The small memory foot print and the low computational cost of the proposed technique make it suitable for person re-identification in large scale surveillance applications.

2011 Fifth ACM/IEEE International Conference on Distributed Smart Cameras, 2011

We propose two novel techniques for automatic summarisation of lengthy surveillance videos, based... more We propose two novel techniques for automatic summarisation of lengthy surveillance videos, based on selection of frames containing scenes most informative for rapid perusal and interpretation by humans. In contrast to other video summarisation methods, the proposed methods explicitly focus on foreground objects, via edge histogram descriptor and a localised foreground information quantity (entropy) measurement. Frames are iteratively pruned until a preset summarisation rate is reached. Experiments on the publicly available CAVIAR dataset, as well as our own dataset focused on people walking through natural choke points (such as doors), suggest that the proposed method obtains considerably better results than methods based on optical flow, entropy differences and colour spatial distribution characteristics.

2009 Digital Image Computing: Techniques and Applications, 2009

ABSTRACT The inherent nature of image and video and its multi-dimension data space makes its proc... more ABSTRACT The inherent nature of image and video and its multi-dimension data space makes its processing and interpretation a very complex task, normally requiring considerable processing power. Moreover, understanding the meaning of video content and storing it in a fast searchable and readable form, requires taking advantage of image processing methods, which when running them on a video stream per query, would not be cost effective, and in some cases is quite impossible due to time restrictions. Hence, to speed up the search process, storing video and its extracted meta-data together is desired. The storage model itself is one of the challenges in this context, as based on the current CCTV technology; it is estimated to require a petabyte size data management system. This estimate however, is expected to grow rapidly as current advances in video recording devices are leading to higher resolution sensors, and larger frame size. On the other hand, the increasing demand for object tracking on video streams has invoked the research on content-based image retrieval (CBIR) and content-based video retrieval (CBVR). In this paper, we present the design and implementation of a framework and a data model for CCTV surveillance videos on RDBMS which provides the functions of a surveillance monitoring system, with a tagging structure for event detection. On account of some recent results, we believe this is a promising direction for surveillance video search in comparison to the existing solutions.

18th International Conference on Pattern Recognition (ICPR'06), 2006

Extracting human postural information from video sequences has proved a difficult research questi... more Extracting human postural information from video sequences has proved a difficult research question. The most successful approaches to date have been based on particle filtering, whereby the underlying probability distribution is approximated by a set of particles. The shape of the underlying observational probability distribution plays a significant role in determining the success, both accuracy and efficiency, of any visual tracker. In this paper we compare approaches used by other authors and present a cost path approach which is commonly used in image segmentation problems, however is currently not widely used in tracking applications.

9th Biennial Conference of the Australian Pattern Recognition Society on Digital Image Computing Techniques and Applications (DICTA 2007), 2007

Smart Cameras are becoming more popular in Intelligent Surveillance Systems area. Recognizing fac... more Smart Cameras are becoming more popular in Intelligent Surveillance Systems area. Recognizing faces in a crowd in real-time is a key features which would significantly enhance Intelligent Surveillance Systems. Using a high resolution smart camera as a tool to extract faces that are suitable for face recognition would greatly reduce the computational load on the main processing unit. This processing unit would not be overloaded by the demands of the high data rates required for high resolution video and could be designed solely for face recognition. In this paper we report on a multiple-stage face detection and tracking system that is designed for implementation on the NICTA high resolution (5 MP) smart camera.

2010 11th International Conference on Control Automation Robotics & Vision, 2010

This paper presents a novel generic weak classifier for object detection called &... more This paper presents a novel generic weak classifier for object detection called "Square Patch Feature". The speed and overall performance of a detector utilising Square Patch features in comparison to other weak classifiers shows improvement. Each weak classifier is based on the difference between two or four fixed size square patches in an image. A pre-calculated representation of the image

Proceedings of the Fourth ACM/IEEE International Conference on Distributed Smart Cameras - ICDSC '10, 2010

... The University of Queensland 4072, QLD, Australia and National ICT Australia PO Box 6020, St ... more ... The University of Queensland 4072, QLD, Australia and National ICT Australia PO Box 6020, St Lucia 4067, QLD, Australia yasir@itee.uq ... 1 –6. [15] J. Cho, B. Benson, S. Mirzaei, and R. Kastner, “Parallelized architecture of multiple classifiers for face detection,” in Application ...

IEEE Transactions on Signal Processing, 1992

This paper defines the 3D reconstruction problem as the process of reconstructing a 3D scene from... more This paper defines the 3D reconstruction problem as the process of reconstructing a 3D scene from numerous 2D visual images of that scene. It is well known that this problem is ill-posed, and numerous constraints and assumptions are used in 3D reconstruction algorithms in order to reduce the solution space. Unfortunately, most constraints only work in a certain range of situations and often constraints are built into the most fundamental methods (e.g. Area Based Matching assumes that all the pixels in the window belong to the same object). This paper presents a novel formulation of the 3D reconstruction problem, using a voxel framework and first order logic equations, which does not contain any additional constraints or assumptions. Solving this formulation for a set of input images gives all the possible solutions for that set, rather than picking a solution that is deemed most likely. Using this formulation, this paper studies the problem of uniqueness in 3D reconstruction and how the solution space changes for different configurations of input images. It is found that it is not possible to guarantee a unique solution, no matter how many images are taken of the scene, they're orientation or even how much colour variation is in the scene itself. Results of using the formulation to reconstruct a few small voxel spaces are also presented. They show that the number of solutions is extremely large for even very small voxel spaces (5x5 voxel space gives 10 to 10 7 solutions). This shows the need for constraints to reduce the solution space to a reasonable size. Finally, it is noted that because of the discrete nature of the formulation, the solution space size can be easily calculated, making the formulation a useful tool to numerically evaluate the usefulness of any constraints that are added.

Visual tracking of the human body has attracted increasing attention due to the potential to perf... more Visual tracking of the human body has attracted increasing attention due to the potential to perform high volume low cost analyses of motions in a wide range of applications, including sports training, rehabilitation and security. In this paper we present the development of a visual tracking module for a system aimed to be used as an autonomous instructional aid for amateur golfers. Postural information is captured visually and fused with information from a golf swing analyser mat and both visual and audio feedback given based on the golfers mistakes. Results from the visual tracking module are presented.

We propose what we believe is a new approach to 3D reconstruction through the design of a 3D voxe... more We propose what we believe is a new approach to 3D reconstruction through the design of a 3D voxel volume, such that all the image information and camera geometry are embedded into one feature space. By customising the volume to be suitable for segmentation, the key idea that we propose is the recovery of a 3D scene through the use of globally optimal geodesic active contours. We also present an extension to this idea by proposing the novel design of a 4D voxel volume to analyse the stereo motion problem in multi-view image sequences.

Circuits and Systems, …, May 8, 1989

A procedure is described for the time-frequency analysis of signals, based on time-frequency dist... more A procedure is described for the time-frequency analysis of signals, based on time-frequency distributions (TFDs) and instantaneous frequency (IF) estimation. First, a suitable TFD is used to determine the number of signal components. Then, if the signal is monocomponent, the IF law can be estimated directly. For multicomponent signals, two-dimensional windowing in the time-frequency domain (a form of time-varying filtering) is used to isolate each component; IF estimation is then applied to the individual ...

The Research Network for a Secure Australia (RNSA) is a multidisciplinary collaboration establish... more The Research Network for a Secure Australia (RNSA) is a multidisciplinary collaboration established to strengthen Australia's research capacity for protecting critical infrastructure (CIP) from natural or human caused disasters including terrorist acts. The RNSA facilitates a knowledge-sharing network for research organisations, government and the private sector to develop research tools and methods to mitigate emerging safety and security issues relating to critical infrastructure. World-leaders with extensive national and international linkages in relevant scientific, engineering and technological research will lead this collaboration. The RNSA also organises various activities to foster research collaboration and nurture young investigators.

The huge popularity of Hidden Markov models in pattern recognition is due to the ability to "lear... more The huge popularity of Hidden Markov models in pattern recognition is due to the ability to "learn" model parameters from an observation sequence through Baum-Welch and other re-estimation procedures. In the case of HMM parameter estimation from an ensemble of observation sequences, rather than a single sequence, we require techniques for finding the parameters which maximize the likelihood of the estimated model given the entire set of observation sequences. The importance of this study is that HMMs with parameters estimated from multiple observations are shown to be many orders of magnitude more probable than HMM models learned from any single observation sequence -thus the effectiveness of HMM "learning" is greatly enhanced. In this paper, we present techniques that usually find models significantly more likely than Rabiner's wellknown method on both seen and unseen sequences.

… , 2005. ITCC 2005. …, 2005

Frank Adelstein Dharma P. Agrawal Igor Aizenberg Giovanni Aloisio Kazumaro Aoki Hamid Arabnia Vij... more Frank Adelstein Dharma P. Agrawal Igor Aizenberg Giovanni Aloisio Kazumaro Aoki Hamid Arabnia Vijayan Asari Michail Attalah Robert L. Baber Pascal Bamford Nick Barnes Emad Bataineh Lejla Batina Siddika Berna Ors Guido Bertoni Euro Blasi Rainer Bluemel Luca Breveglieri Constantine Butakoff Greg Byrd Massimo Cafaro Miriam Capretz Gabriele Carteni Jordi Castella-Roca Herwin Chan Pei-Min Chen Alex Chen Chia-Chu Chiang Jagadish Chintala Edward Christensen Chi-Kit Ronald Chung Vaughan Clarkson Pedro Henrique G. Coelho Nedeljko Cvejic ...