Let’s begin with a simple example. How difficult is it sometimes to find lost room keys in a messy and untidy house? It certainly has happened to the best of us, and till now is still a frustrating experience. But, what if I tell you that a straightforward computer algorithm can help you to find your missing items? This is where object detection using machine learning comes in.
Yes, it is true, technology leaders around the globe are developing software solutions that are powered by machine learning feature detection capabilities. These solutions can quickly detect objects using machine learning algorithms. The example discussed above was a simple one, but the applications of object detection using machine learning, span across several industries from vehicle detection in smart cities to round-the-clock surveillance. Such applications require powerful deep learning algorithms.
The term coined for feature detection using machine learning is image recognition. This article will cover what image recognition is and the use cases of machine learning feature detection.
What is Image Recognition?
Image recognition is the process of identifying and distinguishing image objects within several predefined categories. Thus, image recognition software tools can help users identify what’s depicted in a picture. Computers use machine vision technology in addition to artificial intelligence technology and a camera to realise image recognition.
The image recognition algorithms function requires developers to use comparative 3D models and appearances from several non-identical angles. The algorithms are usually fed with thousands of pre-labelled pictures to help the system mature.
Use Cases Object Detection Using Machine Learning
Now, you have an idea about image recognition and other AI/ML-based possible technologies that are ready to be used for several applications. Let’s now focus on the real-life applications that are powered by these technologies.
1. Visual Listing for Brands
Many brands monitor their social media presence, and brand mentions to learn how their audience perceive, interact, and talk about their brand, by using image recognition tools. The term is coined as “Social Listening” or “Visual Listening”.
It is a fact that more than four out of five images posted on social media with a brand logo do not have a company name mentioned in the caption. Hence it is image recognition softwares that provide the answer to brands willing to obtain brand listening. The software tools will recognise the brand logo, and deliver insights to the concerned authorities.
A startup named Meerkat conducted an experiment that showed how image recognition could make their visual listening effective, by identifying the logo of a brand. In six months, the startup was analysing tweets and other social media posts that had commonly used words for alcoholic beverages, preferably beer. This could be beer, barbecue, bar, Cerveza etc. They trained their AI-powered systems to detect famous brand logos such as Guinness, Heineken, Corona, Budweiser, and Stella. They used these AI learning systems to enable the analysis of images posted on social media which contained those brand logos.
Meerkat analysed more than one million tweets for six months and found a tiny portion of tweets with brand logos. With all the data that they received from the automatic gathering and analysis, they found great insights.
They compared the number of posts containing logos of each brand with their market share and found that these two parameters were nowhere related. A clear example of this is the Guinness brand which had a very small market share of less than one per cent and showed a comparatively impressive presence on social media with eleven per cent as per the data ingested during the experiment.
Another data analyst extracted the geo-coordinates from almost 73% of the images to assess the brand presence across the globe. With the analysis, they found that Bud Light is the most popular beer brand in the USA, while Heineken is more famous around the world, having their largest shares in the US and UK.
Furthermore, the analysts also analysed the images containing people to identify the gender of consumers. Surprisingly, the difference was minor – 1.34% more men posted their pictures with the drinks.
It was only one of the examples of ‘Visual Listening’, brands use this technique for many other purposes such as calculating ROI from sponsoring sports events or making sure that their logo isn’t being misused or misrepresented.
2. Medical Image Analysis
Healtech software solutions powered by machine learning help radiologists reduce their workload of analysing and interpreting several medical images such as ultrasound scans, CT scans, MRIs, or even x-rays.
Medical imaging produces enormous amounts of visual data. IBM found that many emergency room radiologists are expected to examine 200 cases per day, considering many medical studies contain up to 3,000 images. No wonder medical images contribute to 90 per cent of the entire medical data. IBM sees potential in applying AI/ML technologies to derive analysis from the medical images.
AI-based radiology tools don’t replace clinicians but support their decision-making. They flag acute abnormalities, identify high-risk patients or those needing urgent treatment, so that radiologists can prioritise their worklists.
IBM’s research division in Haifa, Israel, is currently developing an AI/ML-based solution called Cognitive Radiology Assistant, which is a next-gen cognitive assistant for radiologists. The software solution provides support to the clinicians and radiologists, by analysing medical images and combining the insights with the patient’s medical records. The scientists also created a deep neural network that is specialised to identify potentially cancerous breast tissue.
3. Image Recognition for Artworks
In this tech-savvy modern world, even the conventional art galleries are utilising object detection using machine learning technology. There are apps that allow users to capture images of any art piece. Using those images, the apps provide users with details such as the creator, art name, year of creation, physical dimensions, material, description, and most importantly, the selling price and price history.
One such app is Smartify, which feeds the museumgoers’ hunger for knowledge. The app is a guide for dozens of museums around the globe, including some very renowned ones like the Royal Academy of Arts, Louvre in Paris, Amsterdam’s Rijksmuseum, the Metropolitan Museum of Art in New York, Smithsonian National Portrait Gallery in Washington DC, the State Hermitage Museum in Saint Petersburg, and many others.
The app uses image recognition technology to match the scanned artworks against its vast digital database of nearly fifty thousand art pieces as of 2017. Anna Lowe, the co-founder of smartify, explains about the way apps work – “We scan digital images of artworks to create digital fingerprints of them. It means that the digital data shrunk to a set of digital dots and lines.
4. Animal Detection and Measurement
Motion-sensing cameras are widely used in natural habitats to capture vast amounts of data on animals. But manual analysis of each image has been a significant obstacle in harnessing the full potential of this automatically gathered data. Several companies are working on developing machine learning feature detection solutions that are capable of automating animal identification with 96.6% accuracy.
The automation of wildlife data collection and analysis will help many fields of ecology such as zoology, wildlife biology, conservation biology, hunting and more. We, at Nimble AppGenie, are currently building an animal measuring AI system using the Yolo3 model and python development language. We’ll soon publish a case study on this after the successful completion of the project. So stay tuned with us on LinkedIn. Here is the link to our LinkedIn page – Nimble App Genie.
5. Facial Recognition to improve airport check-in experience
Facial Recognition is becoming mainstream in several industries, and the travel industry is not an exception. Airlines and airports have started using facial recognition technology to enhance the check-in and boarding experience for their customers. There are two prime reasons behind the adoption of AI in airports. First is to encourage self-service, and second is to make the airport experience faster and safer. Airlines will achieve improved cost efficiency, as they require less staff interaction with their passengers.
The facial recognition boarding equipment scans passenger’s faces. It compares them against the photos stored in the border controlling agency’s database (for example UK Border Agency) to verify passenger identity and flight information. The photos in the database can be from national IDs, visas, or other documents.
American Airlines, for example, have already started using facial recognition at the boarding gates of Dallas Worth International Airport, Texas, Terminal D. Travellers love to get their face scanned instead of using boarding passes. Although, the passengers are still required to carry their passports and ticket to make it through the security check. The facial recognition biometric is always an option for the travellers, to make their experience at the airport more efficient.
6. Visual Product Search
Whilst having a seamless customer buying experience is on the rise, the boundaries between offline and online shopping have now vanished since the retailers adopted visual search. You might already have used Google Lens app to any object or product. But, retailers like Urban Outfitters are making visual search technology a reality in the retail space by introducing the Scan and Shop feature within their eCommerce app.
Sometimes, customers see a product and want to buy instantly or later, but it becomes a hassle for them to either find the product name or details. Object detection using machine learning addresses this issue and allows customers to scan a product they have found in a magazine, physical store, or have seen someone carrying. A quick capture will provide them with detailed information about the work which they can buy online.
Apps with visual product search capability utilise neural networks. These networks process images captured by the users, and generate object descriptions such as fabric, product type, category, colour, etc. Then the solution engine matches the product characteristics against the items/images in the stock/database by using those corresponding descriptions. Based on the similarity score, the apps present the results.
7. Managing SKUs in a retail store
While buying from supermarkets, customers make crucial buying decisions on the shelves. CPG (Consumer Packaged Goods) companies invest heavily in techniques to develop planograms that are an inseparable part of their ideal store strategy. Keeping track of the shelf state with object detection using machine learning digitises the stores, which allows store managers to stay tuned with the shelf conditions.
In 2018, the IHL Group reported that companies bear a total loss of sales of $1 trillion due to products going out of stock on shelves. The study discovered that more than 20% of Amazon’s North American retail revenue was a result of consumers first trying to buy the same product at a local store, but it was out of stock.
The similar study also found that around 32% of the shoppers encountered empty shelves, and therefore, the retailers and CPG companies have to make sure that their shelves are always stocked with the right product. The stores can easily leverage object detection capabilities by mounting cameras in their stores. Doing so will alert the store managers about every empty shelf. The object detection software solutions are capable of immediately alerting the staff on their smartphone or other handheld devices.
Object detection using machine learning detects SKUs (Stock Keeping Units) by analysing and comparing shelf images with the ideal state. Such neural networks are trained to flag gaps between reference planograms and the actual shelf images. It makes the job of auditor very easy, by providing them with real-time feedback on their handheld devices, so that they can take appropriate action immediately.
The need for object detection using machine learning is at an all-time high. Companies are already investing millions of dollars to achieve maximum efficiency. Throughout the article, we’ve seen there are several famous use cases of implementing AI/ML for image/object detection.
The upcoming years will bring more possibilities when the deep learning technology will evolve to enough maturity to deliver 100% accurate analysis. Having the experience of deploying object detection using machine learning, Nimble App Genie is the best pick for companies who want to opt-in for implementing AI and ML technology. Contact us here.