OK, admit it: You uploaded a photo to Microsoft’s How Old Am I? website when it went viral.
If you did, you are not alone. Microsoft sent emails to 50 people asking them to test it. But within two days, 35,000 people had tried it, mostly because it wasn’t terribly accurate and the results were often funny.
The demo was built from Microsoft’s new Face Detection API, part of a gallery of new software tools that generally are grouped into what’s called machine learning. In other words, the algorithm looking at all those faces and trying to guess their age was getting better with each try, at least in theory.
These types of programs and algorithms are important because they are opening up new realms for data analysis: photos and video.
Data, Data Everywhere
YouTube gets 100 hours of new content uploaded every minute. With the advent of cell phone cameras, GoPros, and drone cameras – not to mention traffic cameras, security cameras, and CCTV, video is everywhere.
All that video contains scads of data, but until very recently, the problem was that it required an actual person to watch it in order to analyze it – to see what’s happening, the location where it was filmed, and who is in it. But new technologies have increased our ability to analyze so-called unstructured data such as email conversations, social media posts, video content, photos, voice recordings, sounds, etc.
Combining this messy and complex data with other more traditional data is where a lot of the value lies. Many companies are starting to use big data analytics to complement their traditional data analysis in order to get richer and improved insights and make smarter decisions.
Video Data in Business
The technology to richly analyze video data is still in its nascent stages, but it’s already changing the world.
In days gone by, companies might have video recorded their premises or retail store and kept the recordings for a week or so before recording over them. Now, some of the larger data-savvy retailers are keeping all their CCTV camera footage and analyzing it to study how people walk through the shops, where they stop, what they look at and for how long, so they can make alterations to offers and boost sales.
Some retailers are even using face recognition software, so it probably won’t be long before a combination of data sources such as CCTV camera footage, loyalty card information, and face recognition software will see us being welcomed to a store on our smart phones and directed to particular special offers or promotions based on our previous buying habits.
Assisting in Disaster Recovery
More seriously, when typhoon Hiayan hit the Philippines in 2013, more than 6,000 people were killed and 1.1 million homes were damaged or destroyed. In the UK, a team of volunteers started creating a vital map of the damaged areas using just social media.
Because it is now very common for people to share their experiences as they happen in almost real time, photos, tweets (#Hiayan) and videos about the disaster were being posted on social media. In the aftermath of Hiayan, the volunteers were receiving on average one million photos, messages, tweets, videos, etc., every day.
After filtering the millions of messages using artificial intelligence to pick out the ones that could be important, the volunteers then made an assessment of what they saw. For example, for a photograph they would be asked, “How much damage do you see?” and they simply needed to click the appropriate button: none, mild, or severe. Each piece of data (picture, video, or message) was then assessed by between three to five different people to make sure the assessment was consistent and, therefore, probably accurate.
By pinpointing where the data was coming from in the Philippines (using GPS sensors in the photos or through the text), the volunteers then created an online map, not just of the disaster zone but of the needs in each area.
That meant that when the disaster relief effort arrived in the Philippines, relief workers didn’t need to waste days trying to determine what was happening and which areas were hit the worst. They already knew from the map – created by people halfway around the world – who needed water, who needed food, where the bodies were, where people had been displaced, where the most damage was, and what hospitals were least damaged. This enabled them to better help the injured.
How cool is that?
In the Bay Area in California, a former Google engineer is re-imagining education, with a focus on using technology to improve outcomes.
In the classrooms, students are video and audio recorded constantly, and they are even testing wearables to track students’ movements. Right now, it’s mostly used for health and safety concerns, but the hope is that, eventually, the data will be used to improve teacher performance and assess student mastery.
Of course, collecting data on students with wearables raises privacy concerns. As with most big data technologies, privacy is the most pressing concern for many. As technology becomes cheaper, easier to use, and more ubiquitous, the possibility that we will be video recorded almost everywhere we go becomes more and more realistic.
The question then becomes one of balancing the value that video analytics can provide against the privacy concerns of individuals. Who holds the trump card when these interests collide is yet to be determined.
How do you feel about video analytics? Is your company or field using video to its benefit? Or do you feel like it’s a step closer to a surveillance state? I’d love to hear your thoughts in the comments below.
Subscribe to Data Informed for the latest information and news on big data and analytics for the enterprise.