Martin J. Johnson

Empirical evaluation of a new structure for AdaBoost

We propose a mixed structure to form cascades for AdaBoost classifiers, where parallel strong classifiers are trained for each layer. The structure allows for rapid training and guarantees high hit rates without changing the original threshold. We implemented and tested the approach for two datasets from UCI [1], and compared results of binary classifiers using three different structures: standard AdaBoost, a cascade classifier with threshold adjustments, and the proposed structure.

Performance characteristics of a cost-effective medium-sized beowulf cluster supercomputer

This paper presents some performance results obtained from a new Beowulf cluster, the Helix, built at Massey University, Auckland funded by the Allan Wilson Center for Evolutionary Ecology. Issues concerning network latency and the effect of the switching fabric and network topology on performance are discussed. In order to assess how the system performed using the message passing interface (MPI), two test suites (mpptest and jumpshot) were used to provide a comprehensive network performance analysis. The performance of an older fast-ethernet/single processor based cluster is compared to the new Gigabit/ SMP cluster.

Operating System Virtualization to support E-learning with Affective Intelligent Tutoring Systems

This paper introduces an operating system virtual machine platform for deploying affective intelligent tutoring systems in an e-iearning environment. Managed e-learning environments are often web based and rely on a browser to minimize configuration on the client machine, however even in these scenarios it is not unusual to have very specific requirements of the client machine, browser type and version, Javatrade virtual machine version and even operating system. In the case of affective intelligent tutoring systems there are additional requirements associated with the affective sensors, camera, bio-mouse etc. An operating system virtual machine approach allows the software stack of the client to be pre-configured and distributed in one shot. This minimizes the client side configuration easing the adoption of the technology. This paper identifies some of the key performance issues associated with this approach.

Comparing Intra- and Inter-Processor Parallelism on Multi-Core CellBE Processors for Scientific Simulations

The Cell Broadband Engine (Cell BE) multi-core processor from the STI consortium of Sony, Toshiba and IBM is a powerful but complex processing device that has attracted much attention since its inclusion in Sony PlayStation (PS3) gaming consoles. We report on some performance experiments using the multicore Synergistic Processing Elements (SPE) concurrency capabilities of this chip. We compare performance and software implementation issues with conventional cluster computing techniques such as message-passing, in exploiting clusters of Cell BE processors for scientific simulations. We discuss performance and user programming issues for some hybrid solutions on clustered PS3 computers running Linux.

Empirical measurement of instruction level parallelism for four generations of ARM CPUs

Parallel computing at all levels is becoming important in all devices and not least in mobile and embedded systems. Many wireless, mobile and deployable devices make use of the ARM CPU and its variants. We report on investigations into measuring instruction level parallelism on the ARM processor and on characterising the fine grained parallelism of four generations of ARM cores. We discuss the implications for future applications that use such devices with parallelism and for the future of mobile computing device architectures.

Gaining colour stability in live image capturing

Digital colour cameras are dramatically falling in price, making them affordable for ubiquitous appliances in many applications. An attempt to use colour information reveals a significant problem that usually escapes our awareness. Due to the adaptive nature of the human visual system in most cases we do not recognise most changes in illumination characteristics, a camera however will measure scenes under changing illumination differently. Attempts to deduce object colour from the images will need to cope with the influence of the illumination and the cameras characteristics. Furthermore, a large variety of colour spaces are available to describe colour. Differences between them and their fitness to quantify colour are discussed. This paper tries to establish a basic understanding of the intricacies behind the processes involved in capturing images and recognising colour- from light as a stimulus to the sensed colour values in cameras. The goal is to outline a novel approach fusing common industrial best practices with dynamic adaptation capabilities needed for robustly measuring colour using cameras in real-time. First positive results towards improving colour based reasoning on adaptable colour spaces are stated as an outlook for further development directions.

A unified architecture for the detection and classification of license plates

A method is presented for the detection and classification of New Zealand license plates in real time. The classifier and detector both use a convolutional network which can efficiently be applied to images and is trained using gradient-based learning. The detector has an error rate of less than one percent for individual characters and can find multiple plates in a single image. The classifier has an error rate of less than two percent. The complete system runs at more than 15 frames per second.

Real time pipeline profile extraction using recursive filtering and circle location

This paper describes a pipeline profiling tool which uses a laser to build a 3D model of the inside of a pipe. A robot travels down the pipe finding the profile in real time using a laser and a camera. The laser is used to draw a line around the edge of the pipe; this line is then extracted from acquired images using recursive Gaussian filtering. Three methods were examined for fitting a circle to the profile, two variations of the Hough transform and a random sampling method. The random sampling method was found to be the most flexible and efficient. The extracted profile points were used to construct a 3D model of a pipe section. This model can be viewed using a model viewer written using OpenGL. Results are presented showing the system to be effective and robust.

Halide Vectorization for Android Photography Applications — A Case Study

We evaluate the vector performance of the Halide domain-specific language for a computational photography application targeted at Android devices. Our application has existing implementations in C++ and ARM NEON and these are used as a baseline for performance comparisons with Halide. We give a very brief introduction to Halide concepts and describe the structure of our application. We describe how each stage of the algorithm is implemented and measure the performance of our Halide implementation on 3 popular vector architectures, x86-64 AVX2, ARMv7 NEON and ARMv8.

Real-time segmentation for baggage tracking on a cost effective embedded platform

This paper describes segmentation and tracking parts of a machine vision based airport baggage tracking system. A simplified codebook based background subtraction method is used to segment the bag from a semi-static background. Morphological processing using an integral image is used to filter the foreground mask and the bag location is found using statistical methods. The system was implemented on a cost effective embedded processor and runs in real time at 30 fps. Five ARM based embedded platforms are evaluated and it is shown that all of them are capable of the required performance.