Table of Contents
Computer vision is the most technologically mature field in modern artificial intelligence. This is about to translate into enormous commercial value creation.
The deep learning revolution has its roots in computer vision. At the now-historic 2012 ImageNet competition, Geoff Hinton and team debuted a neural network—a novel architecture at the time—whose performance eclipsed all previous efforts at computer-based image recognition. The era of deep learning was born, with computer vision as its original use case. In the decade since, computer vision capabilities have raced forward at a breathtaking pace.
To put it simply, computer vision is the automation of human sight. Sight is mankind’s most important sense; it underlies much of human life and economic activity. The ability to automate it therefore opens up massive market opportunities across every sector of the economy.
(To be sure, other areas of AI—natural language processing, for instance—have also become increasingly powerful in recent years. But core technology breakthroughs in NLP have come more recently, and as a result NLP remains more nascent from a product and commercial perspective.)
The first wave of entrepreneurial activity in modern computer vision centered on autonomous vehicles. Several startup success stories in that field, including computer vision pioneer Mobileye’s $15.3 billion sale to Intel in 2018, highlight the technology’s power to transform markets and unlock massive economic value.
Today, computer vision is finding applications across every sector of the economy. From agriculture to retail, from insurance to construction, entrepreneurs are applying computer vision to a wide range of industry-specific use cases with compelling economic upside.
Expect to see many computer vision startups among the next generation of “unicorns.” A crop of high-growth computer vision companies is nearing an inflection point, poised to break out to commercial scale and mainstream prominence. It is an exciting and pivotal time in the technology’s journey from research to market.
Agriculture is one of the largest and most important industries in the world. Decisions about how, when and what to farm remain highly underoptimized and imprecise today. An opportunity exists to dramatically improve the food production process using visual data and machine learning.
Based on aerial imagery collected via satellites, drones or planes, computer vision systems can empower growers with real-time insights to optimize their chemical inputs, improve their farming operations and increase their yields.
For instance, image-based analytics can determine which crops would benefit from more or less irrigation, where pipe leaks or pressure failures are adversely affecting crop growth, which areas require more or less fertilization, which fields have suboptimal pest and disease control measures, and so forth. AI systems can make these determinations far more efficiently, reliably and scalably than can humans alone.
Promising ag-tech startups pursuing these opportunities include Ceres Imaging, Prospera, Sentera and Hummingbird Technologies.
There are a number of high-impact use cases for computer vision in retail.
Perhaps the most compelling of these opportunities is checkout-free shopping. The concept is both futuristic and elegantly simple: once a store has been equipped with the necessary sensors and computer vision systems, shoppers can enter, pick up the items they want to purchase, walk out, and receive an automated receipt for their visit without waiting in line.
As with many innovations in retail, Amazon pioneered checkout-free shopping with its Amazon Go program, launched in 2016. A handful of startups is pursuing this opportunity today, including Standard Cognition, Grabango, Trigo Vision, Zippin and AiFi. Standard Cognition, the most well-funded of these competitors, announced a $150 million financing round from SoftBank’s Vision Fund earlier this month.
“The in-person shopping experience will change forever now that computers can see,” said Grabango CEO Will Glaser. “Computer vision systems like Grabango’s detect every product that goes into your cart, so there is no need to re-itemize them at the end of a shopping trip. You just grab, go, and get on with your day.”
In addition to an improved customer experience, checkout-free shopping will enable retailers to reduce labor costs and combat shrinkage.
Inventory management is another important computer vision application in retail. Optimizing product mix on shelves and ensuring that aisles remain stocked throughout the day is a convoluted, dynamic challenge for retailers. Retailers lose many billions of dollars in revenue each year to out-of-stock shelves. Focal Systems is one interesting startup applying computer vision to automate inventory management and reduce out-of-stocks.
The insurance business depends heavily on the visual assessment of assets: to accurately price and underwrite policies, for instance, as well as to determine the extent of damage after an accident for claims purposes. As in other industries, computer vision offers an opportunity to carry out this visual analysis faster, cheaper and more accurately than it is done today.
Cape Analytics and Betterview are two startups applying computer vision to property insurance. Using geospatial data, these companies can automatically evaluate what material a building is made out of, what condition the roof is in, what the roof’s square footage is, how much yard debris the property has, how close a structure is to vegetation, and hundreds of other factors that collectively determine the property’s risk profile and the optimal insurance policy pricing.
Computer vision systems can conduct this analysis instantaneously, at scale, based on learnings from decades’ worth of historical data. Compare this to today’s status quo approach of sending a human to manually inspect properties in person, one by one.
Another startup to watch in this category is Tractable, a London-based company that uses computer vision to generate instant damage estimates after car accidents and natural disasters. These AI-driven estimates help accelerate claims processing and reduce human error.
Construction is a massive and historically underdigitized industry. There are numerous opportunities to boost productivity and save costs in construction through the application of computer vision. An active ecosystem of startups has sprung up to pursue these opportunities.
TraceAir uses drones to collect aerial imagery of construction sites, enabling supervisors to remotely monitor projects and track progress over time. Disperse applies computer vision to build interactive “digital twins” of in-progress construction sites. 1build automates cost estimation in construction by applying computer vision to read floor-plan drawings, material schedules and architectural details on blueprints.
“Cost estimates in construction essentially simulate the full construction process,” said 1build CEO Dmitry Alexin. “Machine learning and computer vision allow us to perform this simulation faster and more accurately, giving construction companies atomic-level visibility into their costs.”
Visual monitoring is at the heart of physical security. The most ubiquitous security device is, after all, the camera. A natural opportunity therefore exists to apply computer vision to make physical security more robust and reliable.
A number of startups are deploying computer vision in innovative ways to enhance and automate the physical security sector.
Verkada offers an AI-enabled security system for commercial properties using hardware sensors, computer vision algorithms and an integrated software platform. The company was valued at $1.6 billion last year, making it one of the few computer vision startups to have already achieved unicorn status.
Deep Sentinel has built a similar solution for home security. The company uses a clever human-in-the-loop model to enable human security personnel to remotely intervene in real-time via microphone when the AI system detects a threat.
“Computer vision is changing everything about physical security,” said Deep Sentinel CEO David Sellinger. “Our AI system acts as a tool to reduce distractions, highlight relevant facts and determine which human guards are right for each situation. Our AI is more accurate and fast-responding than any human can be—and it never has a lapse in attention.”
Another area of security to which computer vision can be usefully applied is checkpoint security: for instance, at airports, live events and government buildings. Fatigued and inattentive human personnel often miss threats at these checkpoints. Computer vision can be applied to camera or X-ray feeds to automatically detect dangerous items with substantially greater accuracy and reliability than a human, improving public safety.
Synapse Technology, one promising startup developing computer vision solutions for checkpoint security, was acquired last year by Palantir.
It is important to note that the application of computer vision in security contexts can and sometimes does cross the line from safety-promoting monitoring to overly intrusive surveillance. Governmental use of facial recognition technology to track and monitor citizens has prompted widespread backlash around the world. In China, computer vision has reportedly been deployed in efforts to suppress Uyghurs, a minority ethnic group.
As with any powerful technology, computer vision can be used in harmful as well as in value-creating ways. It is incumbent upon regulators, businesses and individuals to ensure that society marshals this technology as responsibly as possible.