Mar 31 2023
It’s been 110 years since Frank Gilbreth first used film cameras to improve processes, 36 years since Kei Abe taught me how to do it with VHS, and six years since my last post about videos. Yet the number of manufacturing companies leveraging this tool remains minuscule. The technical hurdles are long gone, but the human ones are still in place. Managers must make it a priority to make the work easier for operators, and operators need to trust management not to use videos against them. My 2nd post on this subject was about management preparation. It’s doable, if not easy.
This post focuses on the easy part: technical issues. This is not to ignore the hard stuff but to address a few recent challenges. Smartphone cameras, by default, now generate high-definition video files so large that they impede analysis. It is, therefore, a good idea to work with lower resolutions. The software to help with time studies based on video has evolved too, and I am including introductions to a few currently available packages. Readers are invited to share their experiences with these or other tools.
- Precision of Timestamps
- Software for Video Analysis
- Software products
A 17-minute video recording with a recent iPhone at its default settings takes up 2.25 GB. This is unwieldy even with 2023 technology, particularly if you record 100 videos and want to open more than one at a time.
The highest possible resolution may be desirable to immortalize a wedding or a bar mitzvah but it is overkill for process analysis. The standard resolution for DVDs was 720×480 pixels, in which a 2-hour movie took ~1.3GB. It is perfectly adequate to observe the movements of an operator’s hands.
There are two remedies:
- For existing videos. Most video players let you save open video files in different formats. In VLC, for example, you choose File>Convert & Stream, then click on Customize, set the resolution to Width = 720, Apply, and then Save as File.
- For future videos. On the iPhone, you can go to Settings>Camera and reduce the resolution to “720p at 30 fps.” On Android, you open the camera app, click on the gear symbol, and adjust the Rear- and Front-video sizes.
Precision of Timestamps
When improving manual operations, the smallest unit of time you consider is usually 1 second, but, to aggregate small increments of time, you avoid rounding errors if you start with measurements in fractions of a second. The precision of timestamps you can extract from a video is bounded by its number of frames per second (fps).
The default for smartphone cameras is 24 fps. An iPhone 13 can go up to 60 fps. Of course, if you double the frame rate, you also double the file size. Even with the default 24 fps, however, you should be able to take timestamps to the nearest 42 ms, giving you a more precise count of whole seconds when you add up 20 steps. Timestamps captured from video are also more accurate than those taken with stopwatches, which have at least a 100 ms delay from when the observer’s brain registers the end of a step to when the finger presses the lap button. It also varies between observers and with tiredness for the same observer.
The most common video players, by default, only display time in seconds. VLC has keyboard shortcuts for slow or fast motion, jumping backward or forwards, and advancing frame by frame. By default, it does not show fractions of a second but it has an extension called time_ext that does. Installing it, however, is a challenge for most end users. This is what the bottom left of the screen looks like with times displayed in milliseconds. With a video captured at 30 fps, the milliseconds count jumps by 33 or 34 with every frame.
Software for Video Analysis
You can start analyzing videos with nothing more than the basic players bundled with operating systems and manually enter timestamps, tags, and comments on a spreadsheet. When you start doing it routinely, you feel the need to automate some repetitive tasks:
- Precisely identify breakpoints in slow motion.
- Collect timestamps.
- Use pull-down menus to categorize tasks.
- Enter key points and comments on a form.
- Review all the clips assigned to a category.
- Zoom in on part of the picture.
This gives you use cases to evaluate products.
Categories of Time
Many products offered for manufacturing limit the categorization of time to abstractions like “value-added,” “non-value added,” and “waste.” These categories are also judgmental, which can be a problem with operators’ perception of their own work. Being told that it is “non-value added” or “waste” demoralizes people and makes them nervous about job security.
It’s OK for software to provide default categories, but it should also let you define your own. To be effective, categories should be concrete, specific, objective, and judgment-free, like “reaching,” “walking,” “mounting,” “checking,” etc. They must be categories that operators understand and are comfortable with. They get the idea of reducing the time spent reaching for parts or walking between stations and don’t have a problem with it.
Software Costs And Installed Base
For video analysis, markets like athletics or education are larger than manufacturing and are served by more offerings at lower fees. Given that relevant time categories vary with each sport, products for this market are more likely to let you define your own categories. On the other hand, they may not support the kind of summarization and analysis you need in manufacturing, which means that you can use them for data collection and export the data to other tools to analyze it.
Organizations Supporting the Software
Functionally, my favorite package used to be Anvil. It was offered free by its developer, Michael Kipp, a professor of interactive media at the University of Applied Sciences in Augsburg, Germany. It had practical limitations on, for example, video file formats. These formats are complicated, and you want the software to read the most common ones without complaining. The “new” version offered for the Mac is from August 2017. I downloaded it but couldn’t install it on my 2021 Mac.
Internal Development versus Commercial Software
Developing your own is not advisable for a function as generic as breaking down a video into segments. It would require software skills that are generally unavailable in a manufacturing company, and it would take time better spent directly improving processes.
Commercial software comes with installers for various operating systems, quick start guides, user manuals, cheat sheets, video tutorials, training courses, support chat boxes, user groups, etc., and it is routinely updated. It takes an organization to provide this support, which users fund through fees or by allowing the sale of their data to 3rd parties. Manufacturers who want to keep their data confidential must remember that, if a service is free to them, they are the product.
Market Size and License Fees
Software of equal power and complexity sells for prices that are determined by market size. Revenues need to support the development organization. If there are 100 users, each pays 1% of the total needed; if there are 1 million users, it’s one ppm for each. Video analysis software is more expensive when developed for manufacturing than for athletics because manufacturing is a smaller market than athletics, not because the software is more powerful.
Categories of Software Suppliers
Who you buy software from matters almost as much as what it does. The best customer experience is usually from a supplier with staying power whose main business is the software. An engineering group within a manufacturing company may develop excellent video analysis software, but management would view selling and supporting it as a distraction, and other manufacturers may be wary of buying it from a competitor. It works best when the manufacturing company turns over its homegrown system to a software company for marketing and support.
A consulting firm may also develop software to enhance its services and provide it to clients, but the primary effect is to generate a demand for implementation consulting services that are far more expensive than license fees.
You shouldn’t, however, exclude any option, as the best is not always available, and functional requirements may trump any other consideration.
This is an attempt at a side-by-side comparison of a few video analysis products, based primarily on information the suppliers volunteer on their websites, and with no star rating. Readers are invited to share their experience of these products, or of other products in the same space.
Timer Pro and OTRS10 are designed specifically for time studies in manufacturing; Dartfish, for athletics; Vosaic, for teachers. Another potential supplier, Hapster, uses video and VR technology to facilitate the transfer of know-how from experienced operators to new ones. Hapster has the video acquisition foundation for time studies and intends to provide these functions but hasn’t yet.
|Company Headquarters||Applied Computer Services, 5445 Dtc Pkwy Pe Greenwood Village, CO 80111||Level 1, 45 Greenhill Road, Wayville SA 5034, Australia||Route de la Fonderie 6, 1700 Fribourg, Switzerland||600 P St Ste 200, Lincoln, NE 68508|
|Application||Video time studies and more||Video time studies + training and consulting||Video analysis for sports||Video analysis for teachers|
|Manuf. Time studies||Core||Core||Repurposing||Repurposing|
|Analysis||Built-in||Built-in||Exports to Excel||XML exports|
|Free trial||45 days||None||15 days||30 days|