How Fastprk2 predicts smart parking occupancy based on deep learning methodologies
Barcelona, February 2018
Deep Learning application for a Smart Parking System
Deep Learning is a branch of machine learning that was inspired by nature. It tries to simulate the ability of the brain to recognize, differentiate and learn patterns from data. The algorithms used are mainly based on neural network models where neurons are organized in layers that are stacked. “Deep” thus describes the multiple number of layers that can be stacked on top of one another. Deep Learning models have shown the ability to achieve great prediction results on big data. Sophisticated Deep Learning algorithms are needed in big data analytics to process data in real-time and achieve high accuracy and efficiency.
When applied to IoT environments such as Smart Parking, Deep Learning can improve parking occupancy predictions thus contributing to more efficient parking guidance, proper parking space utilization and better traffic management.
It is estimated that 30% of city traffic is caused by drivers looking for free parking spots. This does not only lead to traffic jams, but causes public health issues due to tons of extra CO2 being added to the air that inhabitants breath in. What if a city was enabled to use parking space occupancy prediction to guide drivers effectively, especially during peak hours such as early mornings and during rush hour.
Fastprk2 uses Deep Learning for parking space prediction
In line with the EU project Fastprk2, the Fastprk2 team has developed predictive models to forecast the occupancy of parking spots. When using the smart parking management system, cities will be able to effectively manage traffic in urban areas, reduce stress for drivers, improve air quality and increase overall citizen satisfaction.
To obtain meaningful results, the developed models have been trained by the Fastprk2 team with over 2 years of real occupancy data from different parking sectors within one test city. Patterns of parking spot occupancy are far from being linear, thus algorithms are needed to capture nonlinear behaviours. To achieve this, the Fastprk2 team opted for Long Short Term Memory Neural Network (LSTM NN), a model commonly used in Deep Learning. LSTMN NN is a type of network capable of learning order dependence in sequence prediction problems and capturing non-linear time dependent patterns in an effective way. Using LSTM NN, predictions can be done by using a multi-step output strategy.
Multi-step output strategy for parking occupancy prediction
When working with time series data, scientists must decide on how many steps ahead in time the final model should predict. Usually, models only offer 1 prediction: the next prediction based on the time scale of our data. However, there are special kinds of models that can give not just 1 forecasting step but also learn the dependency structure between inputs and outputs, and between outputs and outputs, returning an array of forecasting values ordered in step times. In line with Fastprk2, this means that the accuracy level of the parking space occupancy prediction can be maintained over different future consecutive time steps. This will help cities and parking operators to know hours in advance how the parking sector occupancy will be.
Deep Learning will be integrated in the new Fastprk2 software tool which will display the following information in a user-friendly dashboard:
Benefits for cities and parking operators
The Deep Learning approach used by the Fastprk2 team is capable of offering predictions of parking occupancy ahead of time. The advanced Deep Learning algorithms allow for more accurate parking occupancy forecasts which will enable cities and parking operators to improve parking guidance for drivers and minimize time spent looking for a parking slot. This will not only increase citizen satisfaction, but also public health in general.
In 2016, Worldsensing was granted funding by the European Commission for Fastprk2, an initiative to develop the next-generation of parking detection systems equipped with Intelligent Transport Services (ITS) for cities and citizens, all merged in a single mobility platform. The sensors will combine both magnetic and infrared detection technologies and will pioneer the use of deep learning methodologies to increase data accuracy up to at least 95%.
Check out the Fastprk2 website for more details about the project.
The Fastprk2 innovation project has received funding from the European Union’s Horizon 2020 research and innovation program under the grant agreement No 726607.
Worldsensing is a widely recognized global IoT pioneer. Founded in 2008, the Barcelona-based technology provider delivers Operational Intelligence to traditional industries and cities. With over 80 employees and offices in Barcelona, London and Los Angeles, Worldsensing is globally active and has customers in over 50 countries on 5 continents.