During 2020, advanced driver assistance systems (ADAS) will continue their rise as one of the fastest growing technology segments in the automotive market. 

AI in the driving seat

As manufacturers look to increase safety and improve the driving experience, and as the cost of ADAS technology improves, ADAS capabilities such as L2+ conditional hands free and lane centring will achieve greater market penetration in the mid-range segment of automotive (a topic we talked about in more depth in our recent ADAS whitepaper). 

Artificial Intelligence (AI) is one of the most critical components in modern ADAS. Technologies, such as deep learning, vastly outperform traditional software techniques when it comes to the real-time perception and decision-making tasks needed in advanced ADAS systems. 

Many manufacturers are running comprehensive AI development programmes to enable ADAS to perceive the world around them. However, the proving grounds for these ADAS systems can often be in ideal weather locations. But how can we ensure that such vehicles can navigate in the real world where adverse weather is a common occurrence and ADAS sensors are subject to wear and tear?

Our deep learning team rose to this challenge to develop a state-of-the-art AI solution based on cutting-edge research to address the challenges of adverse weather, dirt and sensor damage. The result was a technology we have named SharpWave™, which creates clear, undistorted views of the real-world from damaged or obscured moving images. 

Trained in our data centre using the NVIDIA POD architecture and NetApp storage, the technology has power to see clearly in difficult, unpredictable situations. We’re delighted NVIDIA has now chosen to make a short film on the technology, which has the potential to transform numerous machine vision and imaging and sensing applications, from autonomous driving to empowering healthcare professionals.

To demonstrate the benefits of using this technology in an ADAS image processing workflow, we conducted a small experiment. We took an off-the-shelf vehicle detection AI model and fed in distorted images. We compared the performance of this to the same model processing images where distortion had been reduced by Sharpwave. As you can see in the video above, there is a dramatic improvement in the performance of the vehicle detection algorithm. 

As a NVIDIA deep learning Service Delivery partner, we look forward to helping more clients and industries respond to and capitalise on the disruptive potential of AI in 2020. 

Author
Thomas Carmody
Head of Transport & Infrastructure

Thomas works with automotive manufacturers and technology providers to develop transformative products and services, leveraging the company’s unique combination of world-leading expertise in silicon, sensing, connectivity, AI. 

 

With over 16 years’ experience in the automotive market, Thomas has led the global development of breakthrough automotive technologies. This includes the world’s first fully AEC-Q100 qualified Bluetooth, BLE and Wi-Fi semiconductor devices, which Thomas brought to market in his previous role as Head of Automotive Connectivity at CSR - later acquired by Qualcomm.  

 

Contact Thomas on LinkedIn or email