Artificial Intelligence and Computer Vision Enable the Next Generation of Visual Intelligence Tools

artificial-intelligence-and-computer-vision-enable-the-next-generation-of-visual-intelligence-tools

Computer vision is one of the hottest areas in artificial intelligence (AI) today. Deep learning has enabled computers to understand images better than humans. It is hard to believe that it was only 5 years ago, in 2012, when the first convolutional neural network (CNN) algorithm showed impressive results in an academic research setting. In 2017, just 5 years later, billions of people are using commercially deployed systems that use the same techniques. Mobile phones today can perform face recognition on the fly; our personal photos are automatically cataloged and made searchable based on their contents, rather than the tags; and we can instantly check the ratings of wines on a restaurant wine list. Facebook has announced plans to use the mobile camera as a window to create fascinating augmented and mixed reality worlds using AI and computer vision, which is the next step after applying fun filters to your selfies.

While computer vision continues to make strides in consumer-facing applications, interesting new enterprise-grade business models are emerging, with the technology becoming widely available through open source libraries and pre-trained models that can be used for creating proofs of concept, ultimately leading to production-grade, commercially-scalable solutions. Two use cases are described below in which the innovative use of computer vision and AI is powering the rise of a new kind of visual intelligence engine, both in the sky and on the ground.

Satellite Imagery for Geo-Analytics

Satellite imagery has long been a closed domain with high-resolution image databases only available to a select few companies and organizations, such as weather centers, government agencies, the military, and oil & gas companies. Companies like Digital Globe that deploy imaging satellites have launched an open data initiative called SpaceNet that provides commercial satellite data, labeled and hosted on the cloud platform AWS. This is similar to ImageNet, which is an open database of images for training AI models. This has become a commercial opportunity for companies like Orbital Insight, Spaceknow, Descartes Labs, and RS Metrics that are building solutions that allow anyone to analyze satellite imagery and perform geo-analytics using computer vision and AI. The number of satellites is also increasing at a rapid pace, with Silicon Valley startup Planet recently deploying 88 satellites in a single launch, with plans to have 143 satellites in orbit soon, mapping every corner of the globe daily. The rapid increase in the availability and improvement in the level of detail of satellite imagery and the advancements in AI and computer vision have created unique business models that now offer vertical markets like oil & gas, agriculture, construction, finance & investment, and government solutions that are catered toward analyzing the globe on a scale that has never been done before.

While many of the applications are fairly common like helping farmers with precision agriculture, helping oil & gas companies discover new energy resources, or aiding environment agencies with monitoring the impact of humans, some interesting applications have emerged in the past few years. These are not just new applications, but new business models that provide country-wide, or object-specific analysis of satellite imagery to vertical markets and include the following:

  • Shipping companies and investment firms can now count the number of ships arriving and leaving ports to gauge trade volume of a country.
  • S. retail index for investors, quantifying traffic and parking patterns at the top U.S. retailers across the country.
  • Global oil storage volume that tracks and identifies oil storage tanks across the world, including those not in the public records to provide a more accurate measure of oil volumes.
  • Measuring metal and commodity production by tracking metals stored outside smelters and storage facilities.
  • Tracking a bounded area, with alerts and updates provided when something changes in that specific area (change detection), or getting historical changes for that specific geographical portion. This was typically the domain of defense and security satellites, now being provided to anyone and everyone.
  • Predictive capabilities around any of the features being tracked, whether it is oil reserves, metals, cars parked, etc.

Crowdsourced Mobile Market Research

The smartphone revolution has allowed for a new kind of business to emerge, one that blends the gig economy with smartphones and AI. Crowdsourced market research using mobile devices enables companies to tap into a crowdsourced mobile workforce that can go into the field and capture images of products, places, and objects, and upload these images to a computer vision platform, which categorizes and tags these images, and performs additional analysis informing market research outcomes. Premise is a company that has pioneered the blending of AI and computer vision to mobile crowdsourcing. Although there were several companies offering mobile crowdsourcing, no one had truly used the power of AI-based image recognition in this context before Premise.

Premise offers crowdsourced market research for a variety of industries, including government, retail, finance & investment, and healthcare, among others. Premise has a network of smartphone users spread across the world. Currently, its network of users is located in 200 cities and towns spanning 30 countries. These are essentially gig workers who get paid per project, which usually includes taking pictures in their location of products, pricing, retail stores, political posters, etc., depending on the data that needs to be gathered.

As in the case of satellite-based geo-analytics, crowdsourced mobile market research is not just a novel application on AI and computer vision, but has given rise to a new type of business model. In this model, companies can hire people on the ground in virtually any location to gather ground-based intelligence. In a sense, crowdsourced mobile market research is highly complementary to satellite-based geo-analytics, as there is no limit to how much data can be gleaned from satellite images versus what can be captured on the ground.

Some of Premise’s interesting use cases include:

  • Capturing data on the percentage of un-electrified homes in Sub-Saharan Africa, including the location, roof size, presence of generators, transformers, and other key variables. This was used by a Global 100 company and its investors to monitor the impact of its energy investments.
  • Standard Chartered used Premise to create food price indicators in Nigeria and Ghana, providing a real-time feed on food prices at the shelf level, allowing its clients to make better investment decisions.
  • Identifying the density and location of political posters for election candidates in Brazil. This helped provide a narrative and real-time analysis about how the election campaign was proceeding, complementing and providing an alternative analysis to mainstream news coverage. The platform could identify defaced posters of one of the leading candidates, Dilma Rousseff, helping identify locations where there was possible opposition to her.

A Powerful Tool for the Future of Data Collection and Analysis

AI and computer vision are leading the creation of a powerful, real-time visual intelligence engine that is global in scale and can look down from the skies to map the reality on the ground. While this raises immediate concerns about privacy and the limits of how and where this intelligence is collected, the current use cases look to be within bounds with a positive impact overall. As we struggle with the rise of fake news and the need for more fact-based analysis, any tool that can help both gather and make sense of the data is a step in the right direction.

Comments are closed.