Smart Parking System
System Architecture
Backend
Machine Learning
- The client is a German startup that wanted to build a smart system to find a free parking spot near the user's destination.
- Our solution covered system design, architecture, and development.
- The core of the system is a set of advanced data analysis algorithms for calculating the probability of finding a parking spot, connected with a route optimization system.
- The system was augmented with a Machine Learning model for car crash detection.
Every driver knows the pain of looking for a parking spot in the city center. Whether you go shopping, want to grab a bite, or just drive to your job, finding a free spacer is like tossing a coin - there's no guarantee you'll get the result you're looking for. But what if there was?
That's exactly what our client thought, deciding to develop a parking optimization system—a bold approach to bringing some relief in everyday struggles.
En route
A German startup contracted Sparkbit to develop a standalone navigation system connected with a parking spot searching module and telematics features. The approach was quite unique, as the application incorporated a peer-to-peer-like model. The data from one user is translated into the information presented to other users.
The system was to operate via a smartphone application paired with a BLE (Bluetooth low-energy) device installed in the car. Both used to collect data. The innovation is based on a robust backend system that applied mathematical models and algorithms to perform advanced data analysis, serving as a foundation for core business functionalities.
We prepared a detailed roadmap focusing on the MVP as a foundation for upgrades. Considering the user-based character of the system, our goal was to release it as soon as possible to start gathering data we needed to extend it.
Key components
Within the first version, we put all the main features:
- intelligent route calculation, maximizing the chances to find a free parking spot
- Smart prediction of where a parking spot will become available within the next 60 seconds
- lightweight emergency notifications to recognize a possible accident and notify a designated person
- "find my car" feature
The system was interconnected. BLE and the smartphone were paired to assign each device to a car, serving as a beacon. Recognizing the driver approaching his car allowed the system to determine whether the vehicle was about to free a parking spot to adjust the route for other users heading that direction.
At the same time, it was a way to make the application operable without the need to register. BLE served as an authenticator for the application, allowing to use the full version of the app without giving any personal information, therefore, protecting user privacy and their data. Along with extra functionalities not directly connected with the parking itself, it was a way to acquire as many users as possible to gather data needed for future development.
We’ve developed advanced mathematical models calculating the probability of finding a free parking spot in a particular area integrated with a route optimization system. The system takes seasonality into account, relying on detected user behavior patterns. It recognizes recurring specified events, such as leaving work on workdays that results in freeing parking space at the same time every day.
Remote collisions
One of the features that required an entirely different approach was the crash detection system. After testing various methods, we prototyped a machine learning model that assessed whether the collision happened based on the accelerometer data from the smartphone.
We had a lot of fun during the experiment phase as we took some unorthodox approaches. We performed validation tests using remote-controlled cars that we crashed with each other to determine if the frequency of accelerometer reading provides enough information to detect a crash with a satisfying level of certainty.
Final touch
We built a fully-operatable system with all the required functionalities. The last thing we had to assure was platform scalability. It was initially meant to be deployed in one city, with a possibility of seamless expansion for larger areas supporting an increasing number of drivers.