Overcrowding automatic detection by using video surveillance in public transportation

In this post, we will be focusing on how Deep Learning applied to passenger counting will improve public transport service quality.

But before going into detail, we invite you to read this previous post. It provides information about the role of Artificial Intelligence in public transport and the problem that many operators and transport authorities have: they don’t know the demand.

But what is Deep Learning?

First, we must start by explaining what Deep Learning is:

Artificial Intelligence: it’s the technique that allows machines to mimic human behavior.

Machine Learning: subset of artificial intelligence techniques that, through statistical methods, allows learning from previous experiences.

Deep Learning: subset of Machine Learning that, through neural networks, allows answers to more complex and abstract questions.

Deep learning comes from the Perceptron model (Minsky-Papert, 1969), but it is in recent years that it has become a reality. And why didn’t it start before? Because it is now that, with the advances in technology, sufficiently powerful processors have been created to carry out Deep Learning classifiers.

Classifiers are algorithms that, by entering information about an image, identify to which category or class it belongs among a certain number of them. For example, by entering an image of an animal, the classifier can know if it’s an elephant or a dog.

And how can Deep Learning help operators and public transport authorities?

Images provided by existing video surveillance cameras in vehicles and platforms can be processed by using Deep Learning for passenger counting and tracking, with the aim of detecting overcrowds in real time.

With the occupancy data, operations managers can take corrective action at the moment. For example, close the accesses if the platforms are too full and not letting more people get into the vehicle if it’s above its capacity.

On the other hand, planners can detect trends and optimize the supply/demand fit by improving scheduling and frequencies of vehicles, as well as adding more fleet in the peak occupancy strips.

From Counterest, we believe that Deep Learning will substantially improve the quality of public transport, so we are already working on our Pulse technology.