Bryson R. Payne

Accelerated 2d image processing on GPUs

Graphics processing units (GPUs) in recent years have evolved to become powerful, programmable vector processing units. Furthermore, the maximum processing power of current generation GPUs is roughly four times that of current generation CPUs (central processing units), and that power is doubling approximately every nine months, about twice the rate of Moores law. This research examines the GPUs advantage at performing convolutionbased image processing tasks compared to the CPU. Straight-forward 2D convolutions show up to a 130:1 speedup on the GPU over the CPU, with an average speedup in our tests of 59:1. Over convolutions performed with the highly optimized FFTW routines on the CPU, the GPU showed an average speedup of 18:1 for filter kernel sizes from 3x3 to 29x29.

Improving GPU particle filter with shader model 3.0 for visual tracking

Human-Computer Interaction is evolving towards non-contact devices using perceptual user interfaces. Recent research in human motion analysis and visual object tracking make use of the Particle Filter (PF) framework. The PF algorithm enables the modeling of a stochastic process with an arbitrary probability density function, by approximating it numerically with a set of samples called particles. The DirectX Shader Model is a common framework for accessing graphics hardware features in terms of shading functionality. In particular, Shader Model 3.0 compliant graphics cards must support features such as dynamic branching, longer shader programs and texture lookups from vertex buffers, among others. In this work, we propose new improvements on previous CPU/GPU Particle Filter frameworks [Montemayor et al. 2004; Lanvin et al. 2005]. In particular, we have reduced bandwidth requirements in the data allocation stage using GPU texture reads instead of CPUGPU memory transfers. But more importantly, using new features in Shader Model 3.0 we can move all the previous particle filtering CPU stages to the GPU, keeping all the computation on the video card and avoiding expensive data readback.

Teaching programmable shaders: lightweight versus heavyweight approach

The most exciting recent advance in computer graphics has been the development of programmable Graphics Processing Units (GPUs). We discuss different approaches to and some of the issues involved in teaching the use of GPUs.

Fair Gain Based Dynamic Channel Allocation for Cognitive Radios in Wireless Mesh Networks

Wireless mesh networks have the potential to deliver Internet broadband access, wireless local area network coverage and network connectivity at low costs. The capacity of a wireless mesh network is improved by equipping mesh nodes with multi-radios tuned to non-overlapping channels. By letting these nodes utilize the available channels opportunistically, we increase the utilization of the available bandwidths in the channel space. The essential problem is how to allocate the channels to these multi-radio nodes, especially when they are heterogeneous with diverse transmission types and bandwidths. Most of current work has been based on the objective to achieve maximal total bandwidths. In this paper, we propose a new bipartite-graph based model and design channel allocation algorithms that maximize the minimal channel gain to achieve relative fairness. Our model maps heterogeneous network environment to a weighted graph. We then use augmenting path to update channel allocation status and use canonical form to compare the new status with previous status to achieve better fairness. Evaluations demonstrate that our algorithms improve fairness compared with related algorithms.

Implementation of residue number systems on GPUs

In recent years, the processing power of graphics processing units (GPUs) has greatly exceeded that of desktop central processing units (CPUs). The current NVidia 7800 series GPU boasts a total throughput of over 300 GFLOPs at peak performance. Furthermore, the rate of increase in processing power in GPUs is roughly double that of Moore’s Law for CPUs. Recent research has examined the use of this computational power for other applications from signal processing to bioinformatics and beyond.

A Machine Learning Based Forwarding Algorithm over Cognitive Radios in Wireless Mesh Networks

Wireless Mesh Networks improve their capacities by equipping mesh nodes with multi-radios tuned to non-overlapping channels. Hence the data forwarding between two nodes has multiple selections of links and the bandwidth between the pair of nodes varies dynamically. Under this condition, a mesh node adopts machine learning mechanisms to choose the possible best next hop which has maximum bandwidth when it intends to forward data. In this paper, we present a machine learning based forwarding algorithm to let a forwarding node dynamically select the next hop with highest potential bandwidth capacity to resume communication based on learning algorithm. Key to this strategy is that a node only maintains three past status, and then it is able to learn and predict the potential bandwidth capacities of its links. Then, the node selects the next hop with potential maximal link bandwidth. Moreover, a geometrical based algorithm is developed to let the source node figure out the forwarding region in order to avoid flooding. Simulations demonstrate that our approach significantly speeds up the transmission and outperforms other peer algorithms.

Securing the Internet of Things: Best Practices for Deploying IoT Devices

The Internet of Things (IoT) has brought a wealth of new technologies both in homes and businesses onto IP networks not natively designed to securely support such myriad devices. Networks once hosting only computers and printers now routinely contain payment systems, Wi-Fi and mobile/wearable devices, VoIP phones, vending machines, sensor and alarm systems, servers, security cameras, thermostats, door locks and other building controls, just to name a few. This chapter analyzes current best practices for securing computer networks with special attention to IoT challenges, discusses selected major IoT security incidents, details selected IoT cyber attacks as proofs of concept, and presents a framework for securely deploying IoT devices in the enterprise and at home.

Motivating secure coding practices in a freshman-level programming course

Secure application development is becoming even more critical as the impact of insecure code becomes deeper and more pervasive in our personal and professional lives. The approach described in this paper seeks to motivate computer science students to write secure code almost from the very beginning by focusing on concrete examples of common software vulnerabilities in the second freshman-level programming course. Sample exercises and assignments are given as examples that can be reused in similar courses. While long-term data collection is still ongoing, initial results are promising enough that the method is presented here in detail to support university faculty interested in incorporating lessons and real-world examples in secure app development in their programming courses at any level.

Automatic 3D object reconstruction from a single image

In this paper, we describe a fully automatic, unsupervised system for recognizing and reconstructing various classes of 3D objects in single-object, light-background 2D images, using a three-layer back-propagation neural network for object classification. Our system takes a single 2D image as input and provides a correctly textured 3D VRML, X3D, or WebGL file for two classes of objects: boxes and spheres. Our approach applies edge detection to the 2D input image, finds the center of gravity of the foreground object, and gathers a set of perimeter distances around the center of gravity. These values are passed to a trained back-propagation neural network with 36 input nodes, 100 intermediate nodes, and four output nodes corresponding to four classes of objects: rectangular prisms (boxes), spheres, cylinders, and organic/other. The approach then deconstructs the object in 2D, reconstructs it in 3D and outputs a textured model in VRML, X3D or WebGL.

Face tracking using skin detection and parallel kernel based methods

Computer architecture is evolving rapidly, and the trend is moving from a single, fast execution unit with large memory space to several execution units with small local memory. Nowadays, even consumer-level systems commonly possess multi-core processors. These lower power-consuming, high-performing multi-core systems are progressively replacing high energy-consumption desktop computers with great success, and the industry predicts that future computing systems will benefit even more from this scalable technology. This evolution is highly beneficial for data-parallel programs, where independent data processing takes place. An example of a future platform in current use is the massively parallel modern graphics processing unit (GPU), with lots of execution units, small and fast memory for each processing core, and high memory bandwidth. Using this platform for demonstration purposes, we can test algorithmic approaches that would scale well to new generations of desktop computers. It is interesting to devise new algorithmic approaches now in fields that require computational demanding tasks, in order to adapt them to new, and future, platforms. In this proposal, we use this underlying idea to propose a parallel, scalable face detection system based on a novel combination of template tracking, skin color detection and a particle filter tracking method.