Sanghamitra Deb, Staff Data Scientist, Chegg
Artificial Intelligence is transforming education, healthcare, retail, automobile and most other industries. Machine Vision or Computer Vision is a remarkable part of AI that gives computers the ability to see, and comprehend, similar to a human being. Vision is a more fundamental means of communication compared to language. A picture is worth a thousand words. With the widespread use ofsmartphones, people want to interact with the digital world via images. Instead of typed instructions, computers will receive information from cameras or sensors.For example, we unlock our phone with our face, we want our fitness app to know what food we ate and the calories associated with it and we want our education app to infer what we need to learn without having to type it. Computer Vision gives machines this ability. The past decade has been revolutionary for Computer Vision with the success of deep learning algorithms and advancement in hardware dedicated to graphs and Image Processing. This has led to a huge uptick in investments in this field.
Fig 1: Investment in Computer Vision since 2011
Hardware Landscape
Machine Vision solutions require fast processing of computationally expensive algorithms on large number of images. The processors might also need to be embedded in drones, smart phones and other portable devices. While processing on computers, we have a large range of hardware options: GPU’s, hybrid GPU+CPU and High Performing Cluster (HPCs). High performing GPU’s and CPU’s are readily available from cloud compute providers.Companies are no longer limited by equipment set up costs and can easily start machine vision projects.
Recent advances in embedded vision systems have produced miniature cameras and processors.There is a huge potential for demand in this space. This will be shaped by the evolving imaging capacities of embedded vision.
Algorithmic Developments
Visual understanding is difficult since it requires knowledge beyond the objects present in image; it represents the ability to comprehend actions and goals of the subjects in an image of video. This is very straightforward for humans but immensely complex for an algorithm.
The future of Machine Vision lies in enhancement of processing speed integrated with improved modeling capabilities
These achievements have been possible because of well-timed hardware and algorithmic advancements, and the availability of examples or, what is more commonly known asannotated data in machine learning.
Machine Vision Applications
To understand the utility of Machine Vision let us deep dive into a few industries that are getting transformed with computer vision.
• Automobile Industry: Every year, traffic accidents account for more 2% of deaths and alarge percentage of people also get severely injured due to human error. Currently, machine vision techniques are being used to incorporate safety features that makes sure the driver stays within the lane and are not too close to other cars. There are also features that help with tight parking situations. Autonomous vehicles are expected to use cameras and sensors all around the car and perform multiple tasks simultaneously to operate the vehicle without human intervention.
• Healthcare: Machine vision is being used in healthcare in a number of applications ranging from predicting heartbeat rhythm disorders to tracking blood loss during childbirth. The advantages of machine vision techniques are precisionand timely detection of rare diseases. Precision medicine can prevent unnecessary invasive surgeries or expensive medications. Machines are able to attain precision from several examples that humans might miss due to sensory limitations. Timely detection of fatal diseases such as cancer with pattern recognition employed by machine vision techniques will have a large impact on saving lives. Machine Vision can also provide assistance to doctors by taking over tedious tasks and making them more efficient. The ability to make predictions at an individual level will make personalized medicine achievable.
Machine vision is also fundamentally changing industries such as quality assurance: defects in production line are automatically detected. For retail, customers could potentially walk into a store and pick the item they want and walk out and there will be a seamless transaction performed by an automated system using machine vision. Other industries that are experiencing a transformation are robotics, agriculture, fitness, education, finance, marketing to mention a few.
What does this mean for your business?
The possibilities of machine vision solutions to improve a business can be numerous. It is important to identify the problems that can be solved by machine vision and get stakeholder buy in. The next step is to decide whether to build a solution in-house or to buy services. If the problem can be cast into a standard problem, such as facial recognition or object detection, thenbuying services is a good place to start. If the problem is custom or domain specific, building solutions makes more sense. This requires expertise which can be achieved through in-house training and external hires. Another important part of achieving an accurate model is having examples or annotated data that deep learning models need, to learn patterns. It is important to collect good-quality annotated data by identifying reliable third-party data vendors with track record of reliable subject matter experts.The secret to successfully implement emerging technologies, such as machine vision to add value to a company, lies in its ability to invest in all these components. Research and Development is another important part of success in machine vision since the field is rapidly evolving.
The future of Machine Vision lies in enhancement of processing speed integrated with improved modeling capabilities. An important factor in adapting vision technology lies in ease of use by the end user who could be a factory operator or a medical staff. In the future, building machine vision capabilities will also get easier with availability of higher-levelabstractions of algorithms and accessible solutions.
