Loading...
Please wait, while we are loading the content...
Similar Documents
2.1 Overview of Face Detection 2.1.1 Problem Definition Look at Previous Work in Face Detection in Section 2.3.2 Below. 2.1.3 General Approach to Face Detection
| Content Provider | Semantic Scholar |
|---|---|
| Abstract | Introduction Humans use vision as their primary means for gathering information about, and navigating through, their surroundings. Providing this ability to automated systems would be a large step toward having them operate effectively in our world. There are, however, two major obstacles to automated vision: incomplete human knowledge of how to reliably derive high-level information from a 2-D image; and the computational complexity of image processing and analysis methods. The latter is of primary concern in the research and development of real-time systems. Only recently has the growth in affordable computing power and research into faster techniques allowed some complex vision tasks to move into industrial and consumer applications. Since many of the most compute-intensive image processing operations are also highly parallel they could be accelerated by orders of magnitude using a customized hardware implementation. This is widely recognized in real-time vision research but rarely attempted since the resources required to design custom hardware are usually not available. Instead researchers direct their efforts toward devising vision algorithms that are efficient when implemented using standard processors. They thus avoid due to computational complexity approaches which are not feasible in software but might work well in hardware. Another option not widely known in the vision community to employ programmable hardware as the implementation vehicle. Programmable hardware has already been shown to be a good solution for many signal processing tasks that are similar to machine vision [25][26][27][28], and using a programmable system reduces the time, cost, and expertise required to create a working hardware prototype. It reduces cost by avoiding the enormous expense of chip and board fabrication and by spreading the cost of the system over all of the vision and non-vision applications for which it might be used. It reduces time and expertise requirements by permitting less rigorous design and testing. When using programmable hardware, the cost of fixing a bug after design is the Chapter 1: Introduction 2 minimal penalty of recompiling rather than the enormous expense of refabricating. The cost of a design change is similarly reduced, thus allowing much more experimentation using differing hardware implementations. The goal of this research is to explore the feasibility of programmable hardware as a platform for complex real-time machine vision. We will do this by implementing a complex vision task on the Transmogrifier-2a (TM-2a), a large configurable hardware system [18][19]. The vision problem we will focus on is object detection: the … |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://www.eecg.toronto.edu/~jayar/pubs/theses/Mccready/RMM.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
