Pascal Bornet is a recognized global expert, thought leader and author in the field of Intelligent Automation. CDO at Aera Technology.

getty

Computer vision — “the eyes of the digital workforce” — is one of the four main components of intelligent automation, a collection of technologies that allow machines to automate and augment the work of white-collar professionals and knowledge workers. The other three components are language, thinking/learning and execution (the “glue” that connects the other components, connecting tasks into a pipeline and allowing the machine to interact with the physical world or with other software and hardware).

Computer vision has been advancing rapidly in recent years due to improvements in neural network technology. It allows computers to perceive elements of the physical world and interpret them, which has a variety of business applications.

Retail Store Automation

Retail store automation uses image and video analysis to automate the checkout process and inventory management, provide traffic analytics, respond to trends in customer behavior for targeted marketing and improve surveillance and security.

For example, Amazon Go uses cameras to identify when a customer picks up or reshelves an item and adds or removes it from their bill, without the need for a human checkout assistant. Poly uses existing store security camera footage to automate checkout and restocking, and V-Count uses camera footage to analyze customers’ behavior in order to target them with real-time promotions.

Intelligent Character Recognition

Intelligent character recognition is the more advanced descendant of optical character recognition. ICR is used to extract and interpret the information from scanned documents and forms, such as invoices, contracts or people’s IDs. Like OCR, it uses computer vision to process scanned images and recognize letters and numbers within them. Using natural language processing and machine learning, it can even cope with unstructured documents or those where the information is not always in the same layout or order (such as invoices from different vendors). Where OCR would turn a scanned invoice into an undifferentiated stream of characters, ICR would be able to identify which characters represent the date, the company name, the amount and so on and populate a table or database with them.

How Computer Vision Can Help Your Business

Most tasks that involve recognizing or analyzing visual data can be automated using computer vision, and computer vision solutions are becoming more powerful all the time due to rapid advances in deep learning.

Computer vision tasks are typically trained using supervised learning, which involves feeding the software a large amount of training data along with the corresponding expected output and allowing it to construct and refine a model for your specific use case. The training process can be expensive and time-consuming, but the resulting system is very likely to yield significant gains in speed and convenience and can often improve accuracy and reliability. If you’re interested in venturing into the world of computer vision, here are a few tips on how to get started.

• Identify use cases across the largest scope of divisions or entities within your business. The broader the scope, the higher the potential benefits and the capacity to invest. If your computer vision project is confined to an organizational silo, it will underperform and lead to reductions in scope and funding. Instead, unlock synergies and economies of scale across the organization.

• Build the business case. Take into account estimated costs versus benefits. Prioritize use cases according to the effort of implementing them versus their benefits. A visible, high-impact pilot will help you maintain momentum and build buy-in across the organization.

• Involve the highest level of management. Getting their sponsorship lets you unlock the necessary funding and talent.

• Start building capabilities as soon as possible. Hire or train the right talents early on so that they own the project and know its history. Consider recruiting internally and upskilling. Similarly, identify what data you need for your use cases, and make sure that data is collected, shared and cleaned in good time so that the project is not held up by it.

Implementing computer vision is not purely a technology project; it also needs business and interpersonal talent. Employees whose jobs may be automated or augmented by it should be educated about the benefits it offers them. To help the project succeed, discuss with them about dropping the most boring and repetitive parts of their job and being freed up for the more fulfilling and creative parts.

Computer vision is only one component of intelligent automation. You can unlock much more capability by combining it with the other components, such as language or execution.


Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?