If you said you didn’t know what visual AI is, maybe you don’t realize that most of us use it every day (and into the night) when we unlock our smartphones through facial recognition. This technical achievement we now take for granted uses computer vision algorithms and deep neural networks to save time, improve security, and enable valuable user experience insights. But when applied to use cases beyond smartphones, the opportunities to use visual AI and computer vision technology to secure our infrastructure and facilities, improve the safety of our workers, and increase productivity when our supply chains are at risk, become not just exciting, but vital.
Last week, SparkCognition announced it has entered into a definitive agreement to acquire Integration Wizards, a leader in visual artificial intelligence (AI). With this acquisition, SparkCognition will expand its technology platform to include visual AI (aka computer vision) capabilities, helping our customers improve health, safety, and environmental (HSE) initiatives, optimize productivity, and take advantage of new insights into their business.
Like many domains of artificial intelligence, visual AI overlaps and complements other AI disciplines and use cases. Its particular role within AI is the ability to “see” and derive meaning from what it observes.
In this introductory primer, let’s begin to unpack the key terminology, tasks, and real-world applications of visual AI.
What is computer vision?
Computer vision (CV) is a field of artificial intelligence that enables computers and systems to capture and interpret meaningful information from image and video data. By applying machine learning (ML) models to images, computers can be taught to accurately identify and classify objects and decide the next best action to take based on what they “see.”
CV’s roots can be traced back to the late 1950s as a sub-domain of early artificial intelligence research and interrelated fields of psychology, neurobiology, cybernetics, mathematics, and computer science. Over decades, a number of foundational computer vision algorithms were developed, including the detection of edges, lines and shapes from images, representing objects as composites of smaller structures, optical flow, and motion estimation. By the 1990s, advances in computing hardware enabled major improvements in neural networks and their optimization algorithms, and the advent of the internet gave researchers access to more visual training data than ever before to accelerate the evolution of the field.
What’s the difference between computer vision, machine vision, and visual AI?
Machine vision, a subfield of computer vision, typically concerns industrial applications, wherein hardware and software combine to provide operational instruction to devices based on the images it’s capturing and processing, according to a general workflow, e.g.:
- A machine vision system captures an image with a camera.
- CV algorithms process and interpret the image.
- Based upon the interpretation, it instructs other system components to act accordingly.
Closely related to computer vision and machine vision, visual AI is a discipline of computer science involving the training of machines to understand images and visual data like humans do. The term “visual AI” is often used interchangeably with “computer vision” but can also refer to interrelated technologies like augmented reality (AR) and virtual reality (VR).
How does visual AI / computer vision work?
Early CV methods focused on low-level image processing techniques for explicitly extracting features from an image (e.g., detecting color, textures, edges, corners, shapes, etc.) and building higher-level tasks for object recognition and semantic understanding composed of these lower-level building blocks.
Deep learning (DL) is a form of representational learning that is particularly well-suited for data sources that are inherently hierarchical. Modern CV approaches are dominated by DL models that exploit the hierarchical nature of vision—automatically learning low- to high-level features from which advanced computer vision tasks are composed. DL helps computer vision attain levels of performance and accuracy that now often exceed those of humans.
What can you do with visual AI and CV today?
CV algorithms implement foundational tasks that help industrial customers leverage camera monitoring to transition from reactive to proactive corrective actions, including:
- Object detection
- Object localization
- Object recognition
- Object counting
- Object tracking
- Image restoration
- Image classification
- Image similarity
- Image captioning
- Image generation
- Semantic segmentation
- Instance segmentation
- Keypoint detection
- Pose estimation
- Action recognition
Today’s visual AI use cases are focused on driving operational excellence, cost reduction/avoidance, and product/service improvement, including:
- Optical character recognition (OCR)
- Document analysis
- Machine inspection (defect inspection, quality control, etc.)
- Medical image analysis
- Robotic vision
- Video surveillance
- Safety (e.g. PPE compliance, fire detection, etc.)
- Automotive safety (machine/vehicular object detection/identification/avoidance)
- Fingerprint recognition and biometrics
- Facial recognition
- Human emotion analysis (mood and sentiment)
- Crowd dynamics
- Retail optimization
The market for visual AI and computer vision
With more than 1 billion CCTV cameras worldwide today, novel applications of CV technologies to solve business problems are expected to mature rapidly this decade. Market demand and technology improvements will fuel transformational change across key industry sectors throughout the decade.
Ultimately, expanding SparkCognition’s portfolio with visual AI capabilities will support our mission to deliver world-class AI solutions that allow organizations to solve their most critical problems, applied across many markets, including energy, manufacturing, government, retail, construction, and mining.
For more information about SparkCognition’s latest acquisition, read our press release here, and stay tuned for more visual AI news in the weeks and months ahead.
