3.2.1 Choice of data collection technique

As mentioned previously, there are two schools of thought on the issue on how information about sign language should be gathered, whether via vision-based techniques or device-based techniques.

A device-based technique was selected for the following reasons:

• Cost. The cost of a camera is significant, but more importantly, graphics boards for capturing the information at a reasonable rate are also costly.

• Bandwidth and processing power. A very high-end glove might produce 40 kilobytes per second of data (200 packets arriving per second each packet consisting of 200 bytes). A low-end camera might produce 1 megabyte per second (320 by 200 image sent 5 times a second). Furthermore, each image must be analysed, relevant data extracted and so on. Storage of the data is also a problem. If four people make ten samples of each of 100 signs, then the signs from a glove would take 160 megabytes, the information from a camera would take 4 gigabytes.

• Complexity and space. Setting up a camera can be complex, and factors such as lighting and so on have to be accounted for. Although similar problems arise with a glove, they are not as dramatic in extent.

• Additional complexity of decoding image data. The thesis is only one year long, and there are limits to how much might be accomplished in one year. Significant effort would have to be exhausted simply extracting relevant data from the frames.

Given that a device-based approach was to be used, there were still some considerations that had to be made. Several options were considered.

• Building the glove from scratch, using as many ``off the shelf'' components as possible. This would offer the possibility to design a glove that suited the purposes of signing most ideally. After speaking with others, however, it was decided that this was a thesis in itself, and that it would be difficult to achieve in time, especially since some of the devices (the small strain gauges to be used for flex sensors and abduction sensors and the position tracking devices) were not locally available.
• Using one of the high-end pre-built gloves such as the VPL DataGlove, the VirTex CyberGlove, and the EXOS Dextrous Hand Master (the EXOS DHM less so because of the difficulty in setting it up on a person's hand). These provide very accurate information about hand position. These would have been ideal, except for one slight point --- they are expensive. The gloves in this range typically start at US$5000. Also, using gloves this expensive would violate one of the goals of the project which was to show that a system could be achieved at reasonable cost.
• Using a Mattel PowerGlove. The Mattel PowerGlove, originally designed for use with video games, surprisingly had its origin in VPL. It has, as previously discussed, many design corners cut.

The approach finally adopted to purchase two PowerGloves with PGSIs and then consider various augmentations to the glove to get it ``up-to-scratch'' if necessary. One glove would be kept as backup.