License Plate identification using Google Cloud Vision OCR + Greeting
License Plate result
Sending Messages to Customers after Scanning their License Plate
3D car model & Representative Dashboard with customer info
Representative Dashboard Page
Special Mascot Generation for the Customer based on Car Make and Car Color
How many times have you been to a dealership? How many times have you been treated poorly, perhaps felt a lack of engagement and trustworthiness with them, or even waited a whole day to get a simple auto service done? CloudPlate has arrived. Keep reading :)
What it does
CloudPlate is a revolutionary and personalized way of welcoming customers to one of many AutoNation dealerships and speed up service processes. This weekend, we strived to create a cost-effective intelligent system to talk to customers during their check-in process, scan their cars and license plates, create 3D models of the cars, and provide all this information together with service history and appointment data to AutoNation representatives.
First, let’s say a customer is arriving at an AutoNation dealership for an extensive brake inspection. They will be instantaneously detected by CloudPlate as they enter AutoNation which will detect the client’s car and plate, greet the customer by first and last name. Simultaneously, CloudPlate sends the customer’s and the car’s information to the CloudPlate Main Dashboard which is available to the representatives of that AutoNation. As all of this goes on, the customer driving the car receives a personalized welcome text message based on the color and make of the car!
The dashboard includes a full menu with the customer’s and the car’s full service history. It also includes a 3D model of the customer’s car, which allows AutoNation’s representatives to better and more quickly help clients with their needs.
How CloudPlate was built
One of our challenges was training a model that would work with low-resolution cameras and put less pressure on our cloud servers. Primarily, we focused on that because we would like to make CloudPlate cost effective. After tuning in the images to get our car and license plate models, we managed to get it working with cheap $15 cameras even in low light conditions.
Additionally, the CloudPlate team wanted to focus on personalization to treat customers especially and differently. So, we programmatically and beautifully created a simple dataset of mascots that changes itself according to the customer's car color and car make. This adds a personalized experience that is impressive and appreciated by the customer.
Accomplishments that the CloudPlate team is proud of
The CloudPlate team is really proud to have put everything together! We connected all the cables between every API we used and created, created a clean and easy-to-use front-end that was 100% made from scratch while delivering a product that is extremely cost-effective, created a product that will increase customer engagement with AutoNation, and that was designed to drastically speed up service appointments at dealerships.
What the CloudPlate team learned
We learned that time-management and task distribution while working under pressure is essential to an effective delivery. This weekend, the CloudPlate team members were not only working here, but they also had to manage their time with programming team meetings at UCF and college exams. It was also important to acknowledge that the CloudPlate team also learned a lot about training machine learning models for the right task (even with videos of cars in a parking lot at night!) and carefully crafted smart manipulation techniques.
What's next for CloudPlate
CloudPlate was built to grab the attention of automotive retailers, specifically AutoNation. It is based entirely in what we believe to be the future of customer engagement and personalization using machine learning and low cost equipment. Our main goal is to better engage customers with AutoNation’s brand and facilitate AutoNation’s services by fully managing their facilities, employees, and customers with CloudPlate efficiently and inexpensively.
Currently private repo due to credentials and API keys. We uploaded the code on the zip file submission