profile

Dominic Tarn December 22, 2022

The Most Exciting Applications of Computer Vision: 2023 Update

blog image

Computer Vision (CV) models are already playing a role in numerous sectors, with hundreds of use cases and commercial applications. 2022 saw an increase in the number of CV-based applications, and we expect the same in 2023, and into the future. 

A lot of work, time, skills, and resources have gone into training CV models and other algorithms. We have to remember that computers can’t “see” in the same way we think of human vision, so anything a CV model identifies is a result of a painstaking process to train it. This work is including but not limited to, applying annotations and labels across thousands of image and video-based datasets

In this article, we take a closer look at how computer vision models are used across numerous industries, and what we can expect from 2023 and beyond, with new innovations and applications emerging. 

What is Computer Vision? 

Going back to basics for a moment: “Computer vision is a field of artificial intelligence (AI) that enables computers and systems to derive meaningful information from digital images, videos and other visual data — and take actions or make recommendations based on that information. If AI enables computers to think, computer vision enables them to see, observe and understand”, according to an IBM definition

At the start of any computer vision project, you need imaging or video-based datasets applicable to the intended use case and application. Cleaning the data is also integral to the process, otherwise, you risk giving a CV model unclean data, tainting the results, and wasting time and resources. 

You need a team of annotators and ideally, annotation tools with automation to annotate and label the images or videos.

After the data cleaning, annotation, and labeling, your ML or data science teams need to train one or more CV, AI, or deep learning models to achieve a high accuracy score, any bias is reduced, and it generates the results you need. Only then can the model go from proof of concept (POC) to production, with iterative feedback loops designed to further improve the results it generates. 

Exciting Computer Vision Based Applications in 2022

Computer vision technology isn’t science fiction; it’s already being used across countless real-world use cases and practical, commercial applications. 

According to Forbes, the size of the computer vision market and its use across dozens of industries means it’s currently worth around $48.6 billion and is expected to keep growing. 

Here are 7 sector examples, containing real-world use cases and applications for computer vision technology, out of hundreds that we could have included: 

Insurance technology

One such example is in the insurance sector. Several startups in the insurance technology sector are using computer vision systems and AI to improve the speed and accuracy of insurance claims. 

It’s a proven technology, being deployed across dozens of market-leading multi-billion dollar insurance companies, already responsible for helping 1 million households, resulting in $2 billion in claims processed more quickly. 

When images or videos of damage are processed through a CV model, insurance companies can more accurately determine the cost of any damage and whether the claim is valid, allowing them to process these claims quicker than traditional approaches. In the insurance sector, the main applications for computer vision algorithms are across home and car repair claims. 

Manufacturing

Computer vision algorithms are also being deployed across the manufacturing sector, interconnected with Internet of Things (IoT) devices embedded in production machines. In the last 10 years, there’s been a digital revolution in manufacturing; some are calling this “Factory 3.0”, or even “Factory 4.0”. CV models are playing a role in this ongoing digital transformation. 

In factories, CV models and real-world applications of annotated image and video datasets are being used to: 

  • Implement fully automated production line assembly; 
  • Detecting defects before products leave the factory; 
  • Generating real-time 3D models, using computer vision algorithms, to improve manufacturing processes that humans struggle with (e.g. hyper-detailed manufacturing, and putting together small parts in machines in everything from electronics to oil and gas); 
  • Rotary and laser die cutting: incredibly precise cutting during the manufacturing process; 
  • Predictive maintenance to increase efficiency and reduce downtime; 
  • Improve health and safety and security; 
  • Inventory management, packaging, shipping, and a whole load more!

Sports Analytics

Analytics and various data-centric monitoring solutions have been in use across the sporting sector for over 20 years. More recently, computer vision technology is being used to create a deeper understanding of player movements in games and training. Combining computer vision models with hyper-accurate recording cameras, and other applications is taking sports analytics to the next level. 

AI-powered performance analysis gives coaches a wider range of insights, and these applications can be used across every level and type of professional sport. 

Identity Verification

In most cases, identity verification software has been developed in the US and Western Europe. As a result, the datasets this software and the relevant face recognition models have been trained on are representative of the majority demographics in those countries and regions. 

Numerous studies have been produced on this issue, so we won’t go into the reasons why or what the tech sector as a whole needs to do to counteract and overcome this bias. 

However, one company, Vida, a digital identity, facial recognition, and verification platform that serves customers throughout Southeast Asia, is overcoming this with its own CV-based solution. 

Vida solved this problem by collecting and training models on South East Asian image-based datasets. In particular, over 60,000 images of people across Indonesia, an ethnically diverse and large series of islands, with a population of 276 million. 

Applying annotations and labels to these datasets was crucial. Otherwise, there would be no way to effectively train CV models to produce an accurate, real-time identity verification solution that’s now used by transport providers, banks, fintechs, and ride-hailing/sharing companies across Indonesia. 

Vida uses Encord’s tools to apply labels and annotations and manage a 20-person labeling team. Encord has been instrumental in improving the efficiency, accuracy, and quality of the labeling and annotation of images that have been fed into the CV models that make Vida’s technology work. 

For more information on how computer vision and Encord are being used for identity verification, here’s an article for a more in-depth look

Agriculture

Even in the agricultural sector, robotics, computer vision systems are being used to reduce the production costs of growing plants in greenhouses. 

For growers of tomatoes, peppers, and cucumbers, this is a game-changing innovation that’s helping them to increase yields and profits while reducing waste and selling more affordable, healthier, local produce. AgriTech companies deploy CV models in the growing and monitoring process, ensuring that plants are reaching their ideal yield size so they can be picked at the right time, reducing food waste, growers' and farmers' carbon footprints, and increasing profits. 

Given the challenges of reducing carbon footprints while tackling supply chain bottlenecks, it’s CV-based innovations such as this that will help the agricultural sector grow and prosper in a changing world. 

Energy and Infrastructure 

Another real-world application of computer vision technology is in the energy and infrastructure sector. 

The energy sector incurs massive losses across vast infrastructure networks. The amount of power a nuclear plant or wind farm generates isn’t the same volume of energy that reaches consumers and businesses. Inefficiencies and a crumbling infrastructure cost this sector billions every year. 

One way to change that is to use a data-driven and AI-based approach to infrastructure inspections and asset management decisions. Images and videos of the component parts of infrastructure networks are processed, labeled, annotated, and run through computer vision models to identify where, when, and how repairs and replacements need to be made. 

Inspection costs are reduced, as are asset maintenance and management costs; thereby, actively increasing energy outputs from generation plants, national transmission, and distribution networks. 

6x Faster Model Development with Encord

Encord helps you label your data 6x faster, create active learning pipelines and accelerate your model development. Try it free for 14 days

Medical Use Cases for Computer Vision 

Medical use cases and applications for computer vision techniques are enormous. 

New innovative approaches are emerging every few weeks. Startups are bringing new applications to the table every day. At Encord, we have worked with dozens of healthcare providers and technology companies in this sector to support CV-based innovations. 

Here are a couple of examples: 

Computer vision models are being used to generate secondary insights from medical imaging scans by an abnormality detection company. For example, scans taken of patients in an MRI machine are to detect a specific illness. However, the scans come back clear, and yet the patient is still showing symptoms of something that can’t be identified. 

When a computer vision model is deployed, these annotated medical images can be processed again to search for unrelated and interconnected health issues that wouldn’t have been detected on the first scan. Using computer topography, medical teams can benefit from a “second pair of eyes”, by comparing the scans to thousands of similar

It’s more time and cost-effective to detect medical abnormalities in patients, identify the real problem, and help to save more lives with better treatment plans. 

Another company, in gastrointestinal care (GI), is using computer vision machine learning algorithms in pre-trial screenings of patients for clinical trials. 

This precision GI healthcare provider is using computer vision models to analyze patient and clinical trial imaging datasets to improve diagnosis accuracy. CV models are trained to analyze thousands of images faster, more accurately, and in much greater detail than a team of medical specialists ever could. 

For more information about computer vision use cases in the medical sector, and future applications of computer vision for healthcare, check out this article

Encord has developed our medical imaging dataset annotation software in close collaboration with medical professionals and healthcare data scientists, giving you a powerful automated image annotation suite, fully auditable data, and powerful labeling protocols.

What Does 2023 Have in Store for Computer Vision Applications? 

At Encord, we fully expect computer vision innovation to accelerate in 2023, driving commercial applications and adoption of this technology forward. We expect this to continue across every field CV technology touches, including medical science and treatments, manufacturing, augmented and virtual reality (AR/VR), self-driving cars, robotics, satellite imagery, manufacturing, retail, and analytics. 

Specific innovations we expect to see more of include: 

  • An increase in the use of active learning pipelines. Active learning (AL) is a process whereby computer vision models — through ML and data ops leaders — actively ask for more information (images, videos, etc.), to improve the performance of the model. Using active learning accelerates model development and improves performance, as we explain in more detail in this article
  • Deploying computer vision closer to the source of the data, using Edge AI (or Edge Intelligence). Edge AI, also known as Edge Intelligence, is when data is captured, stored, and processed close to the source of the data (on edge devices). Applying this using computer vision models would involve pulling the image or video datasets into an annotation platform, and then feeding them directly into a CV micro-model or active learning pipeline. Edge computing using AI is a viable and practical use case in several sectors, and could generate enormous real-world value in healthcare, security, manufacturing, and other industries. 
  • Moving to a data-centric computer vision model. CV data scientists are increasingly focusing on improving the outputs of computer vision models from a different perspective: focusing on the data rather than the models. Instead of constantly trying to modify the models and algorithms, with more effort being put into improving the quality of the datasets, such as the labels and annotations, and the process to create and manage those, the outputs from models can improve, dramatically, in some cases.  

In reality, whether these innovations and new applications are big or small, we anticipate new use cases, applications, products, and services to emerge from the innovative world of computer vision research and technology startups.