Computer Vision Application for Retail Industry

Source: Deep Learning on Medium

Retail Pain Point

Retailers everywhere continuously have to deal with a common pain point: keeping track of all the merchandise. This includes accounting for everything that they sell and what they have in store, as well. The latter can be the most challenging because this includes the merchandise in the warehouse, the store backroom, and even on the shelves as well. In order to make business-critical decisions, brands need to stay always up to date on what they have in their possession, how much they have sold, what the competitors have in stock, and other essential data.

Skywell Software, along with Customertimes , a systems integrator, and our research and development partner, decided to alleviate this business pain point by developing new retail technology. We created an AI-powered mobile app that uses computer vision to enable retailers to keep track of all the merchandise they have in stock. This cutting-edge retail software not only provides pinpoint accurate information but also eliminates mistakes due to human error and synchronizes the data with the server with minimal time-lapse.

The Background

Initially, the client came to us with the need to develop a tool that will allow them to collect data from the flow of merchandise effectively and would greatly increase the quality of business-critical decision making.

The information that such retail store software would provide them with includes the overall flow of the merchandise on the shelves. They need to have such information available for their customers’ brands and the competition as well. Now that we know an idea for the product that Customertimes were looking for, let’s take a look at how Skywell Software approached the problem.

Tech Solution

Keeping all the problem-idea insights in mind, Skywell Software team decided to utilize its deep expertise in computer vision and image recognition to create efficient retail software solutions in the industry. Only with the help of these technologies, we could train the system to locate and identify the products on the shelves, visualizing it in AR, quantify these products and send the information to the server in such a way that the server-side could read this information correctly and could present it in a report taking into account all of the client specifications.

Technologies Involved

As you can imagine, creating such an advanced retail image recognition solution involves using a combination of various technologies. In order to actualize the solution, the client chose the EPSON Moverio BT-300 AR-glasses to visualize the reports since they are much lighter than those of the competition, and their cost was much lower as well.

On our end, we used Unity and SDK Vuforia to achieve the most flexible solution possible, i.e., one that can work on both AR-powered glasses as well as on mobile devices. This played a significant role in choosing the tech stack because one of the main requirements was that the solution must support both visualizations mentioned above. But first things first.

Choosing the proper SDK, we narrowed down our decision to Vuforia and Wikitude. However, after conducting extensive research, we discovered that Wikitude could only support visualization on mobile apps, despite their website claiming that headgear was supported as well.

So, to visualize the information in AR, we chose Vuforia basic SDK that allows for quality object tracking and would enable us to fully actualize the inventory management software that we had in mind. SDK Vuforia turned out to be a universal solution for us.

We chose Unity because this engine gives us the ability to adapt our solution for all platforms. As of today, the platform was built for only Android, but it is meant to be deployed for other platforms in the future.

Overcoming Objections

While choosing the right tech stack and further programming, we had to consider the array of requirements for the solution.

For the client, it was essential to recognize the objects that this system was trained on correctly. Important stages in the development were gaining control of the gyroscope and autofocus of the camera, adjusting the transmission and preparation of data.

It was also crucial that everything was done within the short timeframe since they were looking to present it at an industry event. Taking into account all of the complexities involved in computer vision in retail, we were still able to implement this solution in half the time that other companies suggested.

It was also necessary to ensure the flexibility of the team. Since the deadlines were very tight, the team that worked on the project had to be as accessible as possible for the client, flexible as far as implementation was concerned, and quick-reacting in overcoming the difficulties that arose.

The customer paid particular attention to the scalability of the solution. We had to develop it in such a way that it could be easily scaled in the future — so that new modules like Offline / Online E-Signature, Barcode Scanner, Trade Promotion Management Suite could be added, as well as some new features. Of course, this everything needs to work correctly with each other, which creates proper conditions for integration into an existing product without inner conflicts.

To create a universal solution, we had to choose a technology stack that would ensure the retail shelving solution could work on all platforms (iOS, Windows, etc.) and different devices, as we mentioned above. This we achieved by choosing Unity and Vuforia SDK bundles.

Client Risks

Without such a solution, the client would run the risk of the system incorrectly identifying the items on the images and would not be able to visualize the reports in AR. Skywell Software played a critical part in the development of the product. Without our involvement, Customertimes would not have received a ready-made solution for the retail industry within the set timeframe.

Why Skywell Software

The client chose Skywell Software team due to our expertise and insider know-how. This includes timeboxing — we were the ones who took the responsibility to develop the product in half the time that was offered by the others. Such a timeframe allowed Customertimes to present its ready-made product at the industrial event. Besides, we provided our client with timely, accurate reporting of the findings that were discovered during the research. So, all of the added values played a crucial role in the client’s decision to choose Skywell Software as their research & development partner.


Our results were beneficial for both the Customertimes, and retailers:

– For the Customertimes, we provided a bespoke solution built on a system of independent modules with the possibility of universal porting to other platforms. The solution also complements the client’s product and makes it possible to implement the main idea. Due to our development, their product has become a holistic solution that can successfully perform all the necessary functions and can be sold to direct consumers, i.e., retailers.

– For the client’s customers, we created a useful tool that allows them to reduce the labor costs involved in keeping track of merchandise in stores and the warehouse. This amounted to an audit cost decrease of up to 80%. Our solution also took human error out of the equation. The results speak for themselves. After implementing the solution, sales increased by 2–5% due to the timely monitoring of the availability of goods on the shelves and proper layout.

Let’s Work Together

Share your project ideas, and we will provide you with the best computer vision solution.

Estimate Project