Computer vision has been around for a long time. We have been machine-reading typed text from scanned pages, getting information from QR codes and undertaking other forms of pattern matching on PCB assembly lines with automated optical inspection (AOI) systems. However, human operators have always needed to examine the results, which did not improve over time.
Machine learning (ML) and artificial intelligence (AI) are rapidly changing that and giving a significant boost to what we thought was a mature technology. So, depending on who you talk to, the computer vision market size and forecasts may vary from somewhere between explosive growth and exponential growth—the differences in their optimism are mainly in the semantics.
One such analysis in a MarketsandMarkets Research report estimates the global AI in computer vision market growing from $9.28 billion in 2017 to $48.32 billion by the end of 2023 at a pace of 31.65-percent compound annual growth rate (CAGR). Such a positive outlook should come as no surprise when you realize that just as computer vision needs AI, any AI implementation is most likely to need computer vision because vision forms the basis of most of our own perceptive abilities.
Take improving imaging and image-processing hardware and combine it with the software advancements in ML and AI, and you can develop applications that deliver value much greater than the sum of their parts. But visual data is burdensome to cart over the network back and forth between vision hardware at the edge and AI near a far-flung data center, and that is why AI is moving closer to the vision application itself.
You can see such an implementation in the Vision AI Developer’s Kit by Microsoft and Qualcomm, for instance. It is a camera-based kit that combines Azure IoT Edge, Azure ML and optional Cognitive Services from Microsoft with the Qualcomm Vision Intelligence 300 Platform and the Qualcomm Neural Processing SDK for on-device AI. Engineers will be using vision AI-on-the-edge kits like these to jumpstart new applications that are about to take both consumers and industry by storm.
AI-powered machine vision aids human ops
The breadth of applications for vision AI is staggering, but here is a selection of application domains and use cases in which AI is likely to have the greatest impact.
1. Industrial operations
While computer vision has been used in quality assurance (QA) for a long time, machines have required tedious re-programming for each batch to accommodate even minor visual differences. AI and automatic machine learning promise to take away that pain. AI can train on thousands of items without needing breaks or even calling it a day at the end of a shift, thereby speeding up the return on investment (ROI) for factories.
For instance, AI can learn about different types and classifications for fabric defects, apply statistical and Fourier transforms for image analysis to report frequent faults and give an insight on the adjustment of knitting or weaving parameters.
Beyond freeing up humans from repetitive manufacturing and QA tasks for higher value-added work, vision AI can also help in ensuring their safety in industrial and commercial operations. Oil and gas companies like Shell have already started using AI cameras at the edge to monitor equipment and unsafe behavior. This year, Shell has undertaken a pilot program at gas stations in Singapore using cameras running Microsoft Azure IoT Edge to identify and send alerts on potential safety problems, such as smoking on the premises.
2. Logistics
Computer vision to aid logistics is already underway to log movement of goods as well as assess potential and actual damage at various stages in the supply chain. This enables suppliers to send replacements when needed.
Tesco, a U.K.-headquartered multinational grocery and general merchandise retailer, for example, uses in-store cameras and AI to check item availability and alert staff to fill empty shelves and avoid customer disappointment.
Computer vision will, thus, help tackle bottlenecks in the supply chain by taking over some of the manual processes that are prone to human error or delays.
3. Agriculture
Constant visual monitoring of plants and soil on a farm is not feasible without the aid of technology. Companies are beginning to offer AI solutions that use data from static cameras or those mounted on drones to capture images of farms for analysis.
AI algorithms can use deep-learning concepts to accurately identify soil conditions, such as dryness, pests, plant diseases, weed growth or even time to harvest. For instance, a California vineyard could deploy vision AI to monitor grapevine leaves, which can reveal signs of diseases, such as molds and bacteria, or poor nutrition and get immediate advice on pesticide or fertilizer use.
When you consider the significant impact of soil conditions on food security—the USDA estimates the annual cost of soil erosion at about $44 billion—use of computer vision to take timely corrective action appears a necessity.
4. Traffic monitoring
Vision AI is particularly useful in traffic management and control. Traffic flow measurements that include vehicle counting over time, vehicle classification, car breakdowns, accidents and other driving conditions can be used to predict bottlenecks, control traffic lights or call emergency services. What’s more, historical data can be used to better plan maintenance as well as justify road-widening projects.
With suitable vision AI, it is also possible to detect unusual movements, traffic violations, parking or halting vehicles in unauthorized areas and even excessive vehicular emissions and, combined with plate-number recognition, automatically issue fines or alert law enforcement personnel.
But traffic monitoring doesn’t have to be limited to vehicles. The same technology can be applied to crowd control, for instance, in stadiums and railway stations.
5. Smart home
Vision AI can be a game changer in making homes smart. It can help implement multi-factor authentication (MFA) to control access, monitor unusual movement at home entrances and use facial recognition to verify that anyone entering the premises is registered and authorized, such as a family member. IP cameras can also monitor your home for fire, smoke or flooding. You can be alerted to unauthorized entry and other incidents and have emergency services called should you so choose.
Beyond security, computer vision can be used to offer you personalized comfort. A smart-home assistant can greet you when you come home from work, turn on air conditioning to achieve a temperature of your liking, play your favorite music and automatically check and read personal messages and email. It can suggest TV shows currently on by analyzing your previous preferences or suggest the latest books to download on your tablet for some bedtime reading.
No application untouched
It is difficult to pin down the top use cases for vision AI simply because it leaves hardly any application untouched. While human-displacing robots and self-driving cars get all the attention, vision AI is already beginning to quickly and quietly deliver business and social benefits armed with nothing more than immediately available industrial and consumer IP cameras and AI modeling environments, like the Microsoft AI platform.
To realize such applications, learn about the Vision AI Developer Kit from Arrow.com.
