How to Use Computer Vision for Inventory Monitoring in Supply Chain

Nearly every company and consumer has felt the impacts of a disrupted supply chain as a result of COVID-19. With these challenges, inventory management—especially for organizations involved in manufacturing and warehousing—has never been more important.

All organizations have their own objectives for achieving optimal inventory management, but there are three goals that almost every manufacturing company can agree on:  

1. Avoiding stockouts: The most critical goal for warehouses is to prevent stockouts. Stockouts of any input material for critical operations pose a major threat of downtime to the machinery in the manufacturing process and all other dependant processes. This can have a cascading effect on the cost of operations, as the overall cost of downtime can be quite staggering.
2. Accurate tracking of inventory movement: Organizations often employ barcodes, RFID scanning and other technologies for recording and updating inbound (as part of receiving goods in the warehouse) and outbound (as part of shipping out goods to the manufacturing unit or downstream in the supply chain) movement during the stock lifecycle. The recording of inventory is done multiple times across the inventory lifecycle in the warehouse depending on the operation setup. For example, this could happen in the handover between the transporter and storage area or when putting away stock to storage racks or bins.
3. Accurate stock takes: Organizations conduct stock taking multiple times throughout the month, quarter or even on an adhoc basis. This is important to ensure accuracy in stock measurements.

All of these processes outlined above have one requirement in common: a large amount of resource consumption–both in terms of time and manpower for executing inventory-counting activities. The time and cost of manpower can be minimized (up to a certain extent) with better inventory management software combined with advanced hardware like barcode scanners, hand-held terminals and RFID scanners to make the manual process of scanning and recording more efficient. It’s important to note, however, that while these tools are helpful, there may be inaccuracies as technologies are implemented and staff is trained on how to use them. 

Enter Computer Vision

With processing speed becoming more accessible and common, it has opened the door for automated inventory monitoring through the use of computer vision (CV). Fundamentally, this means counting inventory by using machine learning to process images or video feeds. Many companies already use CCTV cameras in their existing security setup, and this input data source is usually readily available to extend the data from a pure security purpose, making it a relatively easy transition.

In thinking about the objectives outlined above, there are clear use cases where computer vision can be particularly helpful to manufacturers:

• Inventory monitoring for raw materials feeding the assembly line: By capturing images at regular intervals from the camera or video feed, this data can be processed with a CV solution to identify inventory levels of different items stored on the rack. Data can be parsed to the IT system controlling the manufacturing process to safeguard against stockouts of any raw materials necessary for production. Alerts can be triggered when inventory levels fall below set thresholds. This could be particularly useful for a regular production warehouse that is responsible for maintaining a continuous supply of material in a production line.
• Inventory counting at storage: A distribution center or a bulk storage area of raw material or finished goods can employ a similar technical approach. CV can be used to count the number of boxes in storage and feed that data at regular intervals (at the beginning or end of the day for instance), thus eliminating the routine stock-taking process.
• Inventory counting during movement: During inbound and outbound movement during the receiving or shipping process, the camera and video feeds can capture images from different angles at entry and exit doors, or at the staging areas between the doors and storage. A CV solution can process this data and help execute automatic counting of inventory, using it for cross-validation with the information in shipping orders or receiving orders. This can help avoid costs associated with heavy penalties of loss of goods due to miscounting.

Getting Started with Computer Vision

Before making the investment into computer vision, a thorough evaluation should be conducted on the feasibility of implementing this technology. Start by assessing your organization’s readiness from an operations and technology perspective. What changes would the operations team need to make to ensure your CV solution is set up for success? For example, is your field of vision camera in a busy “path of movement” that could disrupt the accuracy of the data? Is there a mix of part codes in the warehouse, requiring your operations team to employ part code identification and segregation? From a technology perspective, where would the CV solution be hosted—on premises or on the cloud?

Next, evaluate your costs. To implement CV effectively, you’ll likely need to set up an extra CCTV. Aside from the actual implementation costs of the CV solution, do you need an implementation partner to help you drive the most value from the technology?

Once you’ve evaluated your readiness and calculated your costs, measure these factors against the impact your CV solution could have on savings. Consider those savings in both enabling automatic counting and minimizing mistakes and stockouts, as well as reducing manpower and operation time.  

There are many tools and technologies available in the market at various levels of maturity that can improve inventory management processes. But when implemented well and at the right time, CV has the capability to help manufacturers achieve optimal inventory management today and in the future.