In this guest blog post, Peter Reid, Head of AI at Mint Management Technologies, discusses the capabilities of computer vision and how it's solving problems at factories and warehouses:
Computer vision is a longstanding field of artificial intelligence (AI) that uses machine learning to teach computers how to look at images from a camera and discern what is depicted. This can allow the machines to raise alerts on a particular occurrence or provide item counts or quality information.
In the past, these interventions, counts, or tracking were performed by humans or sensors. Besides experiencing fatigue, humans require good work environments and represent an expensive way to solve many mundane tasks. Sensors can provide excellent input, but they cost a lot to install and maintain and can become clogged by debris or worn out. Many manufacturing companies and factories are experiencing inventory loss and damage because of mismanagement; they need automated solutions to stay afloat in a highly competitive market.
Fortunately, businesses across various industries can now extract meaningful information from visual data, then take action or make recommendations based on that information.
The Mint Vision Platform, for example, offers a cutting-edge use of computer vision in manufacturing, addressing the need for quality control, accident prevention, and predictive capabilities required for business success. It helps manufacturing industries better manage inventory, improve inefficiencies, decrease costs, and forecast demand for stock.
Improving operational efficiency in factories
There is a significant upsurge in using computer vision to track and control the flow of items in a factory or to pick up incidents or faults. One customer with a massive warehouse was, unfortunately, experiencing millions of dollars in losses to “shrinkage,” which is a subtle way of referring to theft or losses within the factory.
To solve this challenge, a design was created with the Mint Vision Platform running at a series of fixed points: the intake or processing sites for each section, near the conveyor belt sections of the factory. At these chokepoints, cameras are used to monitor all boxes coming in. Each camera identifies the box, recognizes the label, and extracts a picture of the label and its number using optical character recognition (OCR). Face recognition is also applied to note the user dealing with that label. Eventually, all this input gets logged centrally and is searchable through Microsoft Azure Cognitive Search.
By scanning the cameras to count the number of packages coming in and knowing where they were last seen and who handled them, this customer can instantly determine where a package is. Moreover, should a package be missing or damaged, a quick search narrows down the window of time that it could have occurred to hours or minutes.
If data is processed over time, manufacturing managers will be able to detect bottlenecks and start predicting stock levels because they understand the sections of the warehouse that operate faster than others and have better control over staff levels, processing times, and procurement.
Safekeeping of vehicles
Customers that manage vehicles that are driven into a parking lot or a parkade can use the camera and visual system to track the vehicles’ locations as well as note who is driving them and who is nearby.
This results in better tracking and reports of vehicle locations and handling. But the aggregate data gleaned from these systems is invaluable: A complete database of how long each vehicle spends in each area, with each person that works on it, can result in a deep analysis of efficiency and predictive algorithms for lot space, parts, and labor.
The use of computer vision is not limited to particular items on a conveyor belt, such as packages, or items in a lot, such as vehicles. The Mint Vision Platform can be trained to recognize items as varied as bottles, fruit, or pine logs. A factory that processes logs for planks, chipboard, and paper was losing money and efficiency because the processing speed varied greatly across days, different conditions, and different teams. While the operators may have all sorts of useful sensors on the conveyor belts to count how many logs are going through specific points at a time, those proved to be ineffective because they got clogged with sawdust quite often.
The Mint Vision Platform can use computer vision to count the number of logs going past specific points. The customer already has cameras at these points, so there is no need for capital outlay. And the cameras are typically mounted relatively high in the factory and are not subject to the same dust and conditions as sensors. Using computer vision, the factory can track the logs, start estimating the size and quality, and introduce efficiencies into the process.
Using data to solve problems
All the mentioned use cases have something in common: They solve an immediate problem by using visual indicators through computer vision to recognize and track items. This not only solves issues that previously had no easy solution; it also accumulates valuable data that enables you to identify patterns and trends that will give your organization a better sense of existing challenges and inefficiencies.
Using this approach, factories benefit from reduced costs, better inventory planning, and better forecasting, and they enjoy increasing return on investment.
One of the critical points of computer vision is versatility: The same solution that counts logs can monitor kilometers of conveyor belts that transport coal. When too much coal in a particular location on the conveyor belt leads to blockages and spillage, early alerts of overload on the conveyor can be pinpointed. This pre-empts the belt system being shut down, saving on delays and demands to be corrected.
The Mint Vision Platform
The Mint Vision Platform operates under Mint Group, an information technology consultancy with a global network. The platform’s target audience is mining and manufacturing, factories and warehouses, and logistics companies that benefit from AI and computer vision solutions. Factories can plan for storage, embrace search capabilities, and use dashboards to generate reports using the Azure-based solution. Implementation of a system like this can bring immediate, long-term relief from difficulties in processing. But the secondary benefit is gathering clean tracking data, which allows advanced reporting and predictive analytics.
The problems that this type of technology solves were thought unsolvable 10 years ago, and it has opened up many possibilities: housing density reviews with drones, factory inspections in hard-to-reach locations, and even underwater investigation. The use of cameras in the manufacturing industry is a classic example, seeing a surge of interest as the technology matures. The possibilities are endless, considering that the feed can come from fixed cameras, images, or even drones that allow for more advanced inspections.
Learn more: Access a video of Reid presenting at the Conext Africa Conference about AI in the IoT world.