Kate Pioryshkina is Ecommerce Technology Observer at Iflexion
Following the all-permeating trend of automation for every possible process, companies are finding new ways to implement computer vision software in manufacturing and retail, among other industries. Let’s look at real-life applications of computer vision in business.
Demand for automation continues to increase in the manufacturing industry, and computer vision is rightfully finding its way into factory-floor processes. Although we are accustomed to the idea of robots doing humans’ work, significant advancement in computer vision and deep learning makes these machines as powerful as never before there.
Deep learning-based machine vision allows robots to make complex decisions about parts classification, allowing for even less human interference and more accuracy. Embedded algorithms also allow machines to accurately identify defects, which is invaluable for cost-efficiency and overall production effectiveness. Moreover, these intelligent devices are constantly advancing their computer vision capabilities as more data is gathered along the way.
Industrial IoT coupled with computer vision is the next transformative power coming to the manufacturing sector. IIoT improves businesses’ visibility over their operations, allowing for remote control and faster decision-making. Instead of sending unsorted, raw data to other locations for further analysis, machine vision-powered robots can now make decisions on the spot.
For example, when it comes to bin picking, slight variations in bar codes or tiny product defects make people much more prone to error. Such mistakes are inherently devastating to businesses, since shipping damaged or wrong products often leads to reputational damage and financial losses. Advanced computer vision machines can solve this problem entirely.
For example, one large glass manufacturer uses the Spyglass Visual Inspection system to better identify glass flaws during production. One of the major bottlenecks in glass fabrication is frequent ‘false alarms’, such as shadow being identified as a crack. Accurate defect identification and reduction in false positives have allowed this glass manufacturer to save over $1M quarterly, according to the Spyglass vendor.
Retail industry giants, be it brick-and-mortar stores or ecommerce platforms, have figured out that collecting and analyzing data is an ultimate key to success. Computer vision adds a new layer to customers’ data and opens up new opportunities to in-store traffic optimization, as well as risk and cost reduction.
For example, Amazon has already reinvented physical shopping experience with its Amazon Go stores, which don’t require customers to go through the often tedious checkout. The company uses computer vision-powered cameras and weight sensors to determine which products customers put in their shopping bags and automatically bills them via their Amazon account. This makes the retail experience smoother, simpler and entirely cashier-less – ‘just walk out’ indeed.
However, improved consumer experience is probably not the main goal here. Computer vision systems can track customers’ paths throughout the store, which brings invaluable insights about their behavior and shopping patterns. For example, such data can be used to better understand which displays attract more attention and help retailers plan store layouts accordingly.
In theory, facial recognition technology, which also falls into the computer vision category, allows for personalization on steroids. To give you an example, facial recognition sensors can already identify people’s gender, age and most importantly, emotions. Such data can be used to adjust in-store promotions on the fly and provide a maximum level of personalization.
Hardly surprising, such applications of facial recognition bring many transparency and privacy concerns. Although this technology is already heavily used by governments, we can’t expect its widespread adoption in the commercial sector just yet.
Retailers also use computer vision to optimize inventory management. Simbe Robotics’ Tally robot is an exciting example of the powerful symbiosis of AI, computer vision and robotics. Tally operates completely on its own, gathering image data and analyzing it to notify store staff about out-of-stock products or incorrect pricing. The robot does inventory monitoring quicker and more efficiently than humans, allowing store employees to focus on other, more complex tasks.
Future outlook, limitations, and one grand question
Currently, most computer vision applications still need to be fed data externally to provide value. The next big step will be the widespread adoption of vision-enabled robots that can train themselves. As soon as this major shift happens, we will see more machines replacing humans, operating and learning on their own.
However, as we entrust machines to make ever more complex decisions, we will inevitably face the ‘black box’ problem of AI. Even the ones who design sophisticated systems for autonomous cars can’t often connect the dots between AI’s reasoning and its actions in each particular case. And when it comes to safety, we better understand the inner workings of AI to maintain control. A few fatal accidents have already been enough to question the technology and raise many public concerns.
The ultimate question of ‘Do I need computer vision for my business’ is probably too broad and too early to answer. Companies are yet to publicize their projects’ ROIs for us to make conclusions. On the other hand, business leaders need to keep their ears to the ground as computer vision-focused startups continue to receive rounds of six-figure funding.