Turning a Kaleidoscope of Images into Precision Vision

Transportation represents an area where Artificial Intelligence (AI) is contributing to major innovations such as the upcoming disruption of autonomous trains and cars.

 

AI refers to software systems that can make decisions that normally require a human level of expertise, such as understanding, judgment and intention.

One of the key areas currently benefiting from AI is Computer Vision with the implementation of Deep Learning, which uses artificial neural networks to mimic the human brain. This technique aims to learn from voluminous existing data. It then applies that knowledge in interpreting automatically new data such as video-protection scenes(1).  

In fact, this powerful technique is about to replace traditional Video Analytics for two reasons. First, VA’s traditional algorithms are skewed by their extreme sensitivity to the context of usage, including such factors as day/night, crowd, or illumination changes.  Second, compared to what AI brings to Computer Vision, their detection capabilities are limited.

In short, AI is giving surveillance cameras brains to match their eyes, letting them analyse live video and helping operators in the control room to increase their ability to make the right decision.

This is good news for railway safety and security, helping operators more easily spot crimes and accidents while offering a range of applications.

Imagine one example: two fuzzy profiles appearing in the foggy night to dart across a rail track a few hundred meters from the train station. Is it a cause of concern?

The operators may soon rely on data driven operation control centres which have ‘learned’ with AI to distinguish unusual situations that could represent risk, from normal ones such as technicians checking a rail switch.

We usually think of surveillance cameras as digital eyes, watching over us. But really, they’re more like portholes: useful only when someone is looking through them. Sometimes that means a human watching live footage, usually from multiple video feeds.

The use of high-performance computing and deep learning systems give operators the ability to analyse the meaning of images based on past video capture and adapt to new circumstances, so that more information means more understanding.

“Behind the buzz words of artificial intelligence and deep learning, a real disruption is going on in Computer Vision. This technological evolution is providing new features with a real benefit for the customer and fewer false alerts. More reliable analysis from more images can be shared with others and so more informed decisions can be made together, in real time.

Stéphanie Joudrier, ICS Security Product Line Manager at Thales

She also notes how smart video content analysis is not limited to security purposes and can help rail transport operators improve their traffic management in order to provide a better passenger experience.

Let’s say the video analysis shows crowds building regularly on one platform of the station. An alert will provide management with the means to rapidly measure, predict, and address traffic needs by adding trains, for example.

Stéphanie Joudrier

So, while it’s dazzling to behold the reflecting images of a kaleidoscope, today’s myriad and voluminous video images are transformed into intelligent choices for rail system management—and that means a better ride for today’s and tomorrow’s passengers.

 

(1) Video surveillance: How technology and the cloud is disrupting the market. IHS Markit 2017