It has been quite a while since I wrote about symbol recognition and automated takeoff. It’s time to check in on how this technology has improved and where it’s going.

Why are we interested in this technology? Well, counting hundreds of symbols in a few seconds sounds like a good reason to me. Let’s define a few things that symbol recognition is not, starting with character recognition.

Most computers use American Standard Code for Information Interchange (ASCII) to identify characters. Everything typed into a computer has an ASCII code. For instance, 65 is the code for capital A. However, a scanned picture or PDF has no ASCII codes. Character-recognition software can recognize a picture of a letter or number and translate it to an ASCII code. If you have a PDF document you want to edit, character-recognition software can turn the entire document into a word processing file that is ready to edit.

Symbol recognition is not quantity extraction. Software is available to extract quantities from properly built computer aided drafting and building information modeling files. Generally, these kinds of files are not available for the hard-bid jobs most of us are estimating.

Symbol recognition is the ability of a computer program to count symbols that match a defined sample. Most often, estimators draw a box around a symbol they want to count, and the software attempts to find all the matching symbols. Unfortunately, the results are not always perfect due to the varying quality of PDF files. To get the best results, it is important to understand how symbol recognition works.

Most paperless takeoff systems prefer to work with raster images, where the images are made up of dots or pixels. PDF files can contain many types of content, including text, fonts, vector graphics and raster images. Since most takeoff systems prefer raster images, they convert the PDF files to TIFF files, which are 100% raster images. Some systems give you a choice to convert or not, while others automatically convert the file type. Another important factor about these files is the resolution, which is expressed as pixels per inch. Higher resolutions create better pictures, which allows the auto-count to get better results.

Once the software has imported the files for a project, auto-count is ready. Draw a box around the sample and start the auto-count. The software will analyze the pixels in your sample and try to find matches in the page being searched. It is likely that your results will not be entirely accurate. Let’s look at why this happens.

There are many things that can affect the count’s accuracy. Remember, the software is trying to match a sample image to other pictures on the drawing. Poor drawing quality, background images, similar images and slight differences in the symbol being searched for can affect the accuracy of the auto-count. Since we understand how the software works, we can use some strategies to make the results more accurate.

Start by picking the cleanest sample you can find. Make sure there are no background images behind it. I recommend picking more complex samples first. For instance, fixture A uses a circle for a symbol, and fixture B uses two concentric circles. If you count fixture A first, the software will also count fixture B since it has a single circle in it.

Most auto-count programs have tolerance or sensitivity settings that control how close to the sample each symbol has to be in order to get counted. This setting will help the software ignore background lines and branch-circuit lines coming from different directions than the sample. Go ahead and run an auto-count and check the results. If the search missed most of the symbols, or counted all sorts of things that are not the specified symbol, undo the count, adjust the tolerance and run the auto-count again. A few minutes spent fine-tuning this setting on the first drawing will yield superior results on the rest.

Some programs allow you to restrict searches to a defined area on the drawing. This is a great tool to deal with different types of 2-by-4-ft. fixtures using the same symbol. I often see drawings where the fixture type is in a note, such as all fixtures in this room are type A. If you restrict your search to that room, you can get perfect results.

Fortunately, the software continues to get more sophisticated. I can now auto-count pages with high accuracy that were impossible to count when I first started using the software. One person told me the technology is in its infancy. I hope it grows up fast.