Automatic Tire Sidewall Reader: Ever wonder who is the best selling tire in your region? Tire market share numbers can be usually found in tire manufacturer’s annual reports but local region data is hard to come by. Market share tire data on first glance might be a “nice to have ” item but the data could be very useful in a hyper local setting. For example, if an area has a size distribution which skews towards certain sizes, dealers equipped with this information are able to prepare & stock more of these sizes.
Currently if you would like to gather intel on tire market share in your area, you have 2 options. One is to physically go to car parks and manually get the information by doing a car park survey. Alternatively you can ask all the tire dealers for their sales data and make a calculated guess. Both methods seem to be equally daunting and difficult.
Table of Contents
Solution
There must be an easier solution to this problem right? We would like to propose an effortless method to collect tire data through the means of an automated tire sidewall reader. The ideas revolve around taking the picture of the tire sidewall during a stationary stop & process the data via modern image analysis techniques. This method exploits the fact that cars usually have to pass through certain entry points where the car would be in a stationary position. This could be at a car park entry or a toll gate paying situation. The camera system will take a picture of the sidewall and the image analysis system will ideally extract information such as the tire size, brand & DOT date from it.
Hardware
The camera of choice in this case was an Android powered Xiaomi Redmi 9T. The smartphone is ideal in this case due to its relatively cheap price and comes with all the necessary hardware (camera + wireless data transmission) which meet our objectives. Xiaomi Redmi 9T’s camera comes with a 8 MP, f/2.2, 120˚ ultrawide settings which would enable us to capture most of the tire sizes in the market with tolerances in mind.
Testing
The picture below shows the schematic setup of the automatic tire sidewall reader system. Using the phone’s built-in near sensor, the capturing of tire images was automatically triggered when a car is detected close by. Due to the wide angle lens, the camera can set at a comfortable distance while still capturing the minute details of the tire. At a camera to tire distance of approximately 50cm , a width of 120cm can be captured.
After some initial set up errors, the system worked relatively well by capturing 536 tire images out of 752 cars passing by. This provided a yield of 71% which can be accepted from our side. Below are some of the pictures taken from the system.
Image Analysis
Right from the get go, the image processing part of the project would be the most crucial step as this would allow tire specifications to be extracted without any labor effort. After some failed attempts to get our own in house image processing to work, we decided to leverage Google’s own Vision AI. By sending the images to Google’s own Vision AI via the standard API process, we are able to retrieve tons of data from the images we have collected.
The above shows an example for a use case whereby the tire size was accurately identified and some portion of tire product & brand text was extracted out correctly. We found that by using Vision AI, we are able to extract something meaningful (tire size, brand, product name or DOT date) 50% of the time. This number is not awesome but could be further enhanced through machine learning for the use cases with have and also taking better quality pictures.
Summary
Overall, it was a really fun project and we have shown that it is possible to automatically capture tire sidewall images and extract useful data out of it. The 50% yield right now is not ideal for commercial usage but some improvements along the way should push this number to a higher value. Who knows, maybe this could be a commercial feature one day in your parking lot 🙂